Paper deep dive
Generative AI and Federated Learning for Intrusion Detection Systems: A Survey
Jiefei Liu, Abu Saleh Md Tayeen, Pratyay Kumar, Qixu Gong, Wenbin Jiang, Huiping Cao, Satyajayant Misra, Jayashree Harikumar
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 96%
Last extracted: 7/5/2026, 9:59:47 AM
Summary
This survey provides a structured review of the integration of Generative AI and Federated Learning (FL) within Intrusion Detection Systems (IDS). It addresses challenges in modern IDS, such as evolving attack behaviors, data scarcity, and privacy constraints. The paper categorizes generative AI applications (Autoencoders, GANs, Diffusion Models, and LLMs) across various IDS tasks like synthetic traffic generation, data augmentation, and anomaly detection. Furthermore, it explores how these generative models can enhance FL-based IDS to handle non-IID data and communication efficiency, while identifying open research challenges in synthetic data quality and domain-specific LLMs.
Entities (10)
Relation Signals (6)
Generative Adversarial Networks → isa → Generative AI
confidence 100% · categorize generative AI applications in IDS according to model families... covering... Generative Adversarial Networks (GANs)
Diffusion Models → isa → Generative AI
confidence 100% · categorize generative AI applications in IDS according to model families... covering... diffusion models
Large Language Models → isa → Generative AI
confidence 100% · categorize generative AI applications in IDS according to model families... covering... Large Language Models (LLMs)
Federated Learning → enables → Intrusion Detection System
confidence 95% · FL enables distributed IDS training without directly sharing local network traffic
Generative AI → supports → Intrusion Detection System
confidence 95% · Generative models can support anomaly detection, synthetic traffic generation, data augmentation, data imputation, adversarial traffic generation, and IDS alert explanation.
NSL-KDD → usedforevaluating → Intrusion Detection System
confidence 90% · Their experiments on NSL-KDD [5] and UNSW-NB15 [6] show that different adversarial attacks affect classifiers differently
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Intrusion Detection Systems (IDSs) are essential for monitoring network traffic and identifying malicious activities in modern cyber-physical, Internet of Things (IoT), enterprise, and distributed network environments. However, developing reliable IDS models remains challenging because attack behaviors evolve over time, realistic datasets are difficult to obtain, traffic records may be incomplete, attack classes are often imbalanced, and privacy constraints limit centralized data collection. Recent advances in generative artificial intelligence (AI) and Federated Learning (FL) provide new opportunities to address these limitations. Generative models can support anomaly detection, synthetic traffic generation, data augmentation, data imputation, adversarial traffic generation, and IDS alert explanation. FL enables distributed IDS training without directly sharing local network traffic, making it suitable for privacy-sensitive and geographically distributed environments. This survey provides a structured review of generative AI and FL techniques for IDS. We first summarize representative IDS research directions, including adversarial machine learning, anomaly-based detection, IoT-oriented IDS, explainable IDS, and benchmark datasets. We then categorize generative AI applications in IDS according to model families and task objectives, covering autoencoder-based models, Generative Adversarial Networks (GANs), diffusion models, and Large Language Models (LLMs). Finally, we review emerging studies that integrate generative AI with FL-based IDS and discuss open challenges, including synthetic data quality, realistic traffic generation, dual-use adversarial risks, non-IID client distributions, communication-efficient model sharing, federated IDS benchmarking, and domain-specific LLMs for network security.
Tags
Links
- Source: https://arxiv.org/abs/2607.01305v1
- Canonical: https://arxiv.org/abs/2607.01305v1
Trouble viewing inline? Open PDF directly →
Full Text
130,998 characters extracted from source content.
Expand or collapse full text
1 Generative AI and Federated Learning for Intrusion Detection Systems: A Survey Jiefei Liu, Abu Saleh Md Tayeen, Pratyay Kumar, Qixu Gong, Wenbin Jiang, Huiping Cao, Satyajayant Misra, Jayashree Harikumar Abstract—Intrusion Detection Systems (IDSs) are essential for monitoring network traffic and identifying malicious activities in modern cyber-physical, Internet of Things (IoT), enterprise, and distributed network environments. However, developing reliable IDS models remains challenging because attack behaviors evolve over time, realistic datasets are difficult to obtain, traffic records may be incomplete, attack classes are often imbalanced, and privacy constraints limit centralized data collection. Recent advances in generative artificial intelligence (AI) and Feder- ated Learning (FL) provide new opportunities to address these limitations. Generative models can support anomaly detection, synthetic traffic generation, data augmentation, data imputation, adversarial traffic generation, and IDS alert explanation. FL enables distributed IDS training without directly sharing local network traffic, making it suitable for privacy-sensitive and geographically distributed environments. This survey provides a structured review of generative AI and FL techniques for IDS. We first summarize represen- tative IDS research directions, including adversarial machine learning, anomaly-based detection, IoT-oriented IDS, explainable IDS, and benchmark datasets. We then categorize generative AI applications in IDS according to model families and task objectives, covering autoencoder-based models, Generative Ad- versarial Networks (GANs), diffusion models, and Large Lan- guage Models (LLMs). Finally, we review emerging studies that integrate generative AI with FL-based IDS and discuss open challenges, including synthetic data quality, realistic traffic generation, dual-use adversarial risks, non-IID client distri- butions, communication-efficient model sharing, federated IDS benchmarking, and domain-specific LLMs for network security. Index Terms—Intrusion Detection System, Generative AI, Fed- erated Learning, Network Security, Synthetic Data Generation, Large Language Model. I. INTRODUCTION Modern computer networks support a wide range of critical services, including cloud computing, Internet of Things (IoT) platforms, industrial control systems, intelligent transportation, financial systems, healthcare infrastructure, and defense appli- cations. As these systems become increasingly connected, they also become more exposed to cyber attacks that can disrupt services, compromise sensitive data, or damage physical in- frastructure. Intrusion Detection Systems (IDSs) are therefore Jiefei Liu, Pratyay Kumar, Qixu Gong, Wenbin Jiang, Huiping Cao, and Satyajayant Misra are with the Department of Computer Science, New Mexico State University, Las Cruces, NM, USA (e-mail:jiefei, pratyay, qixugong, wbjiang, hcao, misra@nmsu.edu). Abu Saleh Md Tayeen is with the University of Hartford, CT, USA (e- mail:tayeen@hartford.edu). Jayashree Harikumar is with DEVCOM Analysis Center, WSMR, NM, USA (e-mail: jayashree.harikumar.civ@army.mil). an essential component of network security. An IDS monitors system activities or network traffic and identifies behaviors that may indicate unauthorized access, malware propagation, denial-of-service attacks, data exfiltration, or other malicious activities. Early IDS research established the foundation for monitoring security-relevant events and detecting abnormal system behaviors [1], [2]. Since then, IDS techniques have evolved from rule-based and signature-based detection toward data-driven approaches based on machine learning (ML) and deep learning (DL). ML-based IDS models can learn complex traffic patterns from historical data and have shown strong per- formance in both binary attack detection and multi-class attack classification. However, their effectiveness strongly depends on the availability, quality, diversity, and representativeness of training data. In practice, network intrusion datasets are often limited, imbalanced, incomplete, outdated, or collected from restricted environments that do not fully reflect real-world network conditions. Generative artificial intelligence (AI) provides a promising direction for addressing these data-related limitations. Gen- erative models learn the underlying distribution or structure of observed data and can produce new samples that resemble the original data. In the IDS domain, generative models have been used for synthetic traffic generation, data augmentation, missing-value imputation, anomaly detection, adversarial traf- fic generation, and explanation of IDS alerts. Representative generative techniques include autoencoders and variational au- toencoders, generative adversarial networks, diffusion models, and large language models. These methods offer new opportu- nities to improve IDS robustness, especially when real attack samples are scarce, minority classes are underrepresented, or data collection is expensive. At the same time, the deployment of IDS models faces increasing privacy and communication constraints. Traditional centralized IDS training requires collecting network traffic from distributed clients or organizations and transferring it to a central server. This strategy can expose sensitive information and is often impractical for privacy-sensitive environments. Federated Learning (FL) addresses this issue by allowing clients to train models locally and share only model updates with a central server [3]. For IDS, FL is particularly attractive because network traffic is naturally distributed across devices, routers, organizations, and geographic regions. However, FL- based IDS also introduces new challenges, including non- independent and identically distributed (non-IID) client data, communication overhead, client heterogeneity, poisoning at- tacks, and limited access to realistic FL-based IDS bench- arXiv:2607.01305v1 [cs.CR] 1 Jul 2026 2 marks. Although IDS, generative AI, and FL have each been stud- ied extensively, their intersection remains fragmented. Existing studies often focus on a specific model family, a single IDS task, or an isolated FL setting. A systematic review is needed to clarify how generative models are used in IDS, how they can support FL-based IDS, and what technical challenges remain unresolved. Motivated by this need, this survey reviews recent progress in generative AI for IDS and generative AI-embedded FL-based IDS. We organize the literature by IDS research problems, generative model families, application objectives, and FL integration strategies, and we further discuss open challenges and future research directions for building reliable, privacy-preserving, and data-efficient IDS models. A. Research Problems Although IDS has been widely studied, the integration of generative AI and Federated Learning (FL) introduces several open research problems that remain scattered across the literature. This survey focuses on the following questions. RQ1: How are generative models used to improve IDS? Generative models have been applied to IDS for multiple purposes, including anomaly detection, synthetic traffic gen- eration, data augmentation, data imputation, adversarial traffic generation, and alert explanation. However, these applications are often studied independently. A systematic review is needed to clarify which generative model families are used, what IDS problems they address, and how their assumptions differ. RQ2: How reliable is synthetic network traffic for IDS training and evaluation?Synthetic data can mitigate limited data availability, class imbalance, and missing records. How- ever, high statistical similarity to real data does not necessarily guarantee realistic network behavior. In IDS, generated traffic should also preserve protocol constraints, temporal depen- dencies, attack semantics, and network-topology relationships. Therefore, evaluating the quality, usefulness, and realism of synthetic IDS data remains a key challenge. RQ3: How can generative AI support privacy-preserving and communication-efficient FL-based IDS?FL allows distributed clients to train IDS models without directly sharing raw traffic data. However, FL-based IDS faces non-IID client distributions, communication overhead, client heterogeneity, and vulnerability to poisoning attacks. Generative AI can potentially address these issues by augmenting local data, improving minority-class representation, reducing data hetero- geneity, or generating privacy-preserving synthetic samples. The effective integration of generative models into FL-based IDS is still an emerging research direction. RQ4: What datasets and benchmarks are available for evaluating IDS, generative IDS, and FL-based IDS? Reliable evaluation depends on realistic and representative datasets. However, real network traffic is often sensitive and difficult to release publicly. Existing IDS datasets are useful but may be outdated, centrally collected, or insufficient for evaluating federated and generative settings. This survey there- fore summarizes commonly used IDS datasets and discusses the need for more realistic benchmarks for generative AI and FL-based IDS. Based on these research problems, this survey reviews IDS background studies, categorizes generative AI techniques for IDS, summarizes existing IDS datasets, and discusses emerging directions in generative AI-embedded FL-based IDS. Section I Introduction Section I-A Research problems Section I-B Contributions Table I Structure of our survey Section I Intrusion detection systems Section I-A Adversarial Machine Learning for IDS Section I-B Anomaly-based IDS Section I-C IDS for IoT Section I-D Explainable IDS (X-IDS) Section I-E Datasets for IDS Section I Generative models Section I-A Variational Autoencoder Section I-B GAN in IDS Section I-C Diffusion in IDS Section I-D LLM in IDS Section IV Generative AI Embedded Federated Learning based Intrusion Detection System Section IV-A VAE embedded FL-IDS Section IV-B GAN embedded FL-IDS Section IV-C Diffusion embedded FL-IDS Section IV-D LLM embedded FL-IDS Section V Conclusion TABLE I: Structure of this survey B. Contributions The main contributions of this survey are summarized as follows: •We provide a structured review of IDS research from the perspective of data-driven security modeling, covering representative directions including adversarial machine learning for IDS, anomaly-based IDS, IoT-oriented IDS, explainable IDS, and IDS benchmark datasets. This back- ground establishes the technical context for understanding why generative AI and Federated Learning (FL) are increasingly important for modern IDS development. •We present a taxonomy of generative AI applications in IDS by organizing existing studies according to both model family and task objective. Specifically, we review autoencoder-based models, Generative Adversarial Net- works (GANs), diffusion models, and Large Language Models (LLMs), and analyze how they are used for 3 anomaly detection, synthetic traffic generation, data aug- mentation, data imputation, adversarial traffic generation, and IDS alert explanation. •We systematically examine the emerging intersection of generative AI and FL-based IDS. Different from con- ventional IDS surveys that study centralized detection models, this survey highlights how generative models can support privacy-preserving and distributed IDS training by addressing non-IID client data, class imbalance, com- munication cost, adversarial robustness, and limited local data availability. •We summarize commonly used IDS datasets and FL- oriented IDS datasets, emphasizing their roles, limita- tions, and suitability for evaluating generative AI and FL-based IDS. This dataset-level discussion helps clar- ify current benchmark gaps, especially the shortage of realistic, topology-aware, and federated IDS datasets. •We identify open challenges and future research direc- tions for generative AI-enabled IDS, including synthetic data reliability, realistic network traffic generation, evalu- ation of generated samples, privacy-preserving data aug- mentation, communication-efficient generative FL, and domain-specific LLMs for network security. C. Survey Methodology To assemble the literature reviewed in this survey, we conducted a keyword-based search across Google Scholar and DBLP, supplemented by manual inspection of recent proceedings and articles from major security, networking, and machine learning venues. Our search combined two sets of terms: generative-model keywords (generative AI, VAE, variational autoencoder, GAN, diffusion model, large language model, LLM) and application keywords (intrusion detection, IDS, network security, federated learning, anomaly detection, synthetic traffic, data augmentation). Candidate papers were then selected primarily by relevance rather than by a fixed time window: a study was included if it applied a generative model family to an IDS task, integrated generative models with FL- based IDS, or provided foundational background (e.g., seminal generative architectures, IDS datasets, or evaluation metrics) needed to interpret these works. We excluded papers that used generative models in unrelated domains without transferable methodology and those that did not provide enough method- ological detail to assess. Because generative AI is a fast- moving field, the large majority of the application-oriented studies we review were published within roughly the last five to seven years, while a smaller set of older references is retained to establish the technical lineage of each model family (for example, the original VAE, GAN, and diffusion formulations) and the early foundations of intrusion detection. This relevance-driven strategy lets the survey track current practice while preserving the conceptual context required to compare model families and identify open challenges. Following this methodology, the remainder of this survey is organized as shown in Table I. Section I first provides the background of IDS by reviewing representative research direc- tions, including adversarial machine learning, anomaly-based detection, IoT-oriented IDS, explainable IDS, and commonly used IDS datasets. Section I then reviews generative AI techniques for IDS and categorizes existing studies by model family, including Variational Autoencoders (VAEs), Gener- ative Adversarial Networks (GANs), diffusion models, and Large Language Models (LLMs). Section IV further discusses the integration of generative AI with Federated Learning (FL)-based IDS, focusing on how different generative models can support distributed, privacy-preserving, and data-efficient IDS training. Finally, Section V concludes the survey and summarizes future research opportunities. I. INTRUSION DETECTION SYSTEMS(IDS) Cybersecurity aims to protect computer systems, networks, and data from unauthorized access, misuse, disruption, and damage. As modern services increasingly depend on inter- connected infrastructures, detecting malicious activities before they cause serious impact has become a core requirement for network defense. Intrusion Detection Systems (IDSs) address this requirement by monitoring system events or network traffic and identifying behaviors that may indicate security violations. The concept of intrusion detection was first intro- duced by James Anderson in the early 1980s [1]. Building on this foundation, Denning proposed one of the earliest functional IDS models, which formalized intrusion detection as the process of monitoring audit records and identifying abnormal or suspicious system behavior [2]. In general, an IDS can be implemented as a software- or hardware-based security mechanism. Its primary goal is to detect potential attacks, policy violations, or abnormal activities and provide alerts that support timely investigation and response. Over the past few decades, IDS research has evolved from rule-based and statistical methods to machine learning and deep learning approaches. Existing IDS studies differ not only in the detection algorithms they use, but also in the attack assumptions, deployment environments, data sources, and explanation mechanisms they consider. Therefore, before reviewing generative AI and FL-based IDS, it is necessary to summarize the main IDS research directions that provide the technical context for this survey. In this section, we review representative IDS-related studies from five perspec- tives. Section I-A discusses adversarial machine learning for IDS, focusing on the vulnerability of ML-based detectors to adversarial manipulation. Section I-B reviews anomaly-based IDS, where attacks are detected by modeling deviations from normal behavior. Section I-C summarizes IDS studies for IoT environments, where resource constraints and heterogeneous devices introduce additional challenges. Section I-D reviews explainable IDS, which aims to make detection results more interpretable for security analysts. Finally, Section I-E dis- cusses commonly used IDS datasets, which are fundamental for training, evaluating, and comparing IDS models. A. Adversarial Machine Learning for IDS Machine learning (ML) and deep learning (DL) models have been widely adopted in IDS because they can learn discrimi- native patterns from network traffic and detect attacks beyond 4 manually defined signatures. However, the use of ML also in- troduces a new attack surface. An adversary may intentionally perturb input traffic features or manipulate training data to mislead the detector, causing malicious traffic to be classified as benign or forcing benign traffic to be reported as malicious. These threats are generally studied under adversarial machine learning, which focuses on understanding the vulnerability of ML models and improving their robustness against adversarial manipulation. For IDS, adversarial robustness is especially important because network attackers can actively adapt their behaviors after observing or probing the detection system. Even small modifications to traffic characteristics may change the predic- tion of an ML-based IDS while preserving the attack objec- tive. Therefore, researchers have investigated both adversarial attacks against IDS models and defense strategies that make IDS models more stable under adversarial conditions. Jmila and Khedher [4] evaluated the vulnerability of IDS models built with shallow ML classifiers, including Deci- sion Tree, Random Forest, and Logistic Regression. Their experiments on NSL-KDD [5] and UNSW-NB15 [6] show that different adversarial attacks affect classifiers differently, indicating that IDS robustness depends on both the attack strategy and the underlying model architecture. Alotaibi and Rassam [7] further surveyed adversarial attacks against ML- based IDS and summarized representative defense strategies. These studies show that adversarial machine learning is an important research direction for IDS, particularly as IDS models become more data-driven and are deployed in adaptive threat environments. B. Anomaly-based IDS IDSs are commonly categorized as signature-based, anomaly-based, or hybrid systems according to their detection strategy. A signature-based IDS detects intrusions by matching observed activities with predefined attack signatures or rules. This approach is effective for known attacks, but it is limited when facing new or evolving threats that do not match existing signatures. In contrast, an anomaly-based IDS first models nor- mal system or network behavior and then identifies activities that deviate from this learned baseline. This capability makes anomaly-based IDS particularly useful for detecting unknown or zero-day attacks, although it may also introduce higher false-positive rates when benign behavior changes over time. Several surveys have reviewed anomaly-based IDS from different perspectives. Yang et al. [8] conducted a systematic review of network IDS studies and analyzed commonly used data processing techniques, evaluation metrics, benchmark datasets, and detection models. Hajj et al. [9] presented a tax- onomy of network attacks and discussed attack tools, relevant detection features, IDS data sources, dataset types, system architectures, and detection modes. They also summarized key challenges that affect the effectiveness of anomaly-based IDS, including dataset quality, feature selection, evaluation consistency, and deployment constraints. These studies show that anomaly-based IDS is a central direction in IDS research because it directly addresses the limitation of signature-based detection under unknown attacks. However, its performance depends heavily on how normal behavior is modeled, how representative the training data are, and how deviations are distinguished from benign traffic variations. C. IDS for IoT The Internet of Things (IoT) connects sensors, actuators, embedded devices, and coordinator nodes to provide net- worked services in domains such as smart homes, healthcare, transportation, and industrial systems. Compared with con- ventional networks, IoT environments are more heterogeneous because devices may differ in hardware capability, operating system, communication protocol, and deployment context. Many IoT devices also have limited computation, memory, and energy resources, which makes it difficult to deploy complex security mechanisms directly on the devices. In addition, vulnerabilities in firmware, hardware, and protocol imple- mentations can expose IoT systems to malware propagation, denial-of-service attacks, spoofing, and unauthorized access. These characteristics make IDS design for IoT different from traditional IDS design. Several studies have reviewed IDS techniques specifically for IoT environments. Kumar et al. [10] provided a taxonomy of ML-based IDS for secure IoT communication and compared different categories according to their advantages, limitations, and resource-related evaluation metrics. They also proposed an IDS model that combines convolutional neural networks (CNNs) with fuzzy rules to improve performance under IoT constraints, including energy consumption and packet delivery ratio. Jayalaxmi et al. [11] analyzed ML- and DL-based IDS and Intrusion Prevention Systems (IPS) for IoT. They further proposed a risk factor analyzer and a hybrid Intrusion Detection and Prevention System (IDPS) framework to ad- dress limitations of purely anomaly-based or signature-based methods. These studies show that IoT-oriented IDS must consider both detection accuracy and deployment constraints. Effective IDS for IoT should be able to detect diverse attacks while remaining lightweight, adaptive to heterogeneous devices, and practical for resource-constrained environments. D. Explainable IDS (X-IDS) Machine learning (ML) and deep learning (DL) models have been widely used in IDS because they can learn complex traffic patterns and improve attack detection performance. However, many DL-based IDS models operate as black-box systems, meaning that their internal decision process is difficult for human analysts to interpret. This lack of transparency limits the ability of administrators and security experts to understand why an alert is generated, identify the root cause of an attack, and determine an appropriate response. Explainable Artificial Intelligence (XAI) addresses this issue by providing interpretable evidence or explanations for model predictions. In the IDS context, explainability is important not only for improving user trust, but also for supporting incident analysis, model debugging, and security decision-making. 5 Several surveys have studied explainable IDS from different perspectives. Moustafa et al. [12] presented a comprehensive survey on XAI methods for cyber defense, with a particular focus on anomaly-based IDS in IoT networks. Their work reviewed studies at the intersection of XAI, anomaly-based intrusion detection, IoT security, summarized open challenges, and future directions for explainable cyber-defense systems. Neupaneet al.[13] proposed a taxonomy of X-IDS techniques and categorized existing methods into white-box and black- box approaches. They also introduced a three-layered X- IDS architecture inspired by the DARPA XAI program [14] and discussed key challenges in developing explainable IDS models. These studies indicate that X-IDS is an important direction for practical IDS deployment. A detection model with high accuracy may still be difficult to use in real security operations if its alerts cannot be interpreted. Therefore, explainable IDS aims to bridge the gap between automated detection and human-centered cyber-defense analysis. E. Datasets for IDS Benchmark datasets are essential for developing, evaluating, and comparing IDS models because they provide shared traffic records, attack labels, and feature representations for reproducible experiments [6], [16], [17], [19]. In IDS research, dataset quality directly affects the reliability of model eval- uation. A dataset should contain representative benign and malicious traffic, diverse attack types, clear labeling rules, and sufficient feature information for downstream detection tasks. However, constructing realistic IDS datasets is difficult because real network traffic may contain sensitive user in- formation, organization-specific configurations, and security- critical infrastructure details. Several studies have reviewed the limitations and design considerations of IDS datasets. Ringet al.[23] analyzed network-based IDS datasets from the perspective of attack scenarios and discussed the relationships among different datasets. Their work also provided recommendations for dataset selection and future dataset construction. Khraisat et al.[24] reviewed IDS datasets together with detection techniques, data collection methods, evaluation practices, and dataset limitations. These studies show that dataset selection is not only an experimental detail but also a key factor that determines whether IDS results can generalize to realistic deployment environments. Table I summarizes commonly used IDS datasets and recent domain-specific datasets. The listed datasets cover tradi- tional network intrusion detection, distributed denial-of-service detection, IoT traffic, in-vehicle network security, and FL- oriented IDS evaluation. The bold datasets, including UNSW- NB15, CICIDS2017, CICDDoS2019, and CICIoT2023, are emphasized because they are widely used in recent IDS studies and provide relatively large-scale traffic records with multiple attack categories. In contrast, datasets such as Car Hacking, CIC IoV, and X-CANIDS are more domain-specific and are mainly designed for vehicle or controller area network security. FLNET2023 is particularly relevant to FL-based Generative AI with IDS Large Language Model (LLM) I-D Generative Diffusion I-C Generative Adversarial Networks (GAN) I-B Variational Autoencoder (VAE) I-A Fig. 1: Generative AI with intrusion detection system IDS because it explicitly supports federated evaluation, while most existing IDS datasets are centrally collected and do not naturally reflect client-level data distribution. Although these datasets have supported substantial IDS research, several limitations remain. Many benchmark datasets are collected in controlled environments and may not fully cap- ture real network topology, user behavior, temporal dynamics, or evolving attack strategies. In addition, most datasets are designed for centralized learning and provide limited support for studying non-independent and identically distributed (non- IID) clients in FL-based IDS. These limitations motivate the development of more realistic, privacy-aware, and federated IDS benchmarks. I. GENERATIVE MODELS Generative AI refers to a class of models that learn the un- derlying structure or probability distribution of observed data and use the learned representation to generate new samples. In the IDS domain, generative models are mainly used to address data-related limitations, including limited attack samples, class imbalance, missing values, and insufficient traffic diversity. They are also used for anomaly detection, adversarial traffic generation, and explanation of IDS outputs. As shown in Figure 1, this survey organizes generative AI techniques for IDS into four representative model families: Variational Au- toencoders (VAEs), Generative Adversarial Networks (GANs), diffusion models, and Large Language Models (LLMs). Section I-A first reviews VAE-based IDS studies. VAEs encode input data into a latent distribution and reconstruct or generate samples from this latent space, making them useful for anomaly detection, data generation, and data augmentation. Section I-B then discusses GAN-based IDS methods. GANs use a generator and a discriminator in an adversarial training process and have been widely applied to synthetic traffic generation, class-imbalance mitigation, and adversarial attack generation. Section I-C reviews diffusion-based methods, which generate data through a denoising process and have recently been extended from image and text generation to tabular data generation. Section I-D discusses LLM-based 6 Dataset NameYearTraining instancesTesting instances# of classes# of FeaturesFL support KDD CUP 99 [15]19991,074,992311,029541No ISCX NSL-KDD [5]20094,898,431311,027442No UNSW-NB15[6]2015447,915-1049No CICIDS2017[16]20172,830,743-884No CSE-CIC-IDS2018 [16]201816,136,255-1276No CICDDoS2019[17]201950,063,112-1184No Car Hacking [18]20208,694,507-512No FLNET2023 [19]20236,807,107-1184Yes CICIoT2023[20]202346,556,613-2539No CIC IoV [21]20241,408,219-612No X-CANIDS [22]2024286,293-5689No TABLE I: Representative IDS datasets. Datasets in bold are widely used benchmark datasets in recent IDS research. IDS applications, including traffic-log analysis, tabular data generation, attack classification, and IDS alert explanation. Network traffic data used in IDS can appear in different formats, including flow-level tabular features, packet-level byte sequences, traffic logs, and raw packet capture (PCAP) files. Therefore, different generative model families can be applied depending on how the traffic is represented. Tabular generative models are suitable for flow-based IDS datasets, sequence and language models can be used for logs or byte- level representations, and image-based generative models can be applied when traffic records are transformed into image- like formats. The following subsections summarize how these generative models have been used in IDS and what technical challenges remain for each model family. A. Variational Autoencoder in IDS Variational Autoencoders (VAEs) are generative neural net- works introduced by Kingma and Welling [25]. A VAE encodes input data into a latent space and reconstructs the input from the learned latent representation. Different from a conventional autoencoder, a VAE models the latent represen- tation as a probability distribution rather than a fixed vector. This probabilistic design allows the model to sample from the latent space and generate new data points that follow the learned data distribution. Therefore, VAEs are suitable for both representation learning and synthetic data generation. In IDS research, VAEs are mainly used for anomaly de- tection, data generation, and data augmentation. For anomaly detection, a VAE learns the distribution of normal or known traffic and identifies samples that are difficult to reconstruct. For data generation and augmentation, a VAE can generate synthetic network traffic samples to improve data diversity, mitigate class imbalance, or support downstream IDS training. This subsection reviews VAE-based IDS studies according to these major application objectives. 1) Anomaly Detection:VAE-based anomaly detection re- lies on the assumption that normal traffic can be reconstructed more accurately than abnormal or unseen traffic. After training, the VAE reconstructs an input sample from its latent represen- tation, and the difference between the original input and the reconstructed output is measured as the reconstruction error. A large reconstruction error indicates that the input does not VAE Anomaly Detection Evaluation Case Studies [26], [27] Performance Metrics [28], [29] Applications Critical Infrastructure [30], [31] Insider Threat Detection [32] Network Traffic Analysis [33], [34] Techniques Latent Space Analysis [35], [36] Reconstruction Error [37], [38] Fig. 2: Variational autoencoder anomaly detection in intrusion detection system follow the learned traffic distribution and may correspond to malicious or abnormal behavior. As shown in Figure 2, this mechanism has been widely used in anomaly-based IDS. Several studies have applied VAEs to network-flow anomaly detection. Zavrak and Iskefiyeli [38] used VAE-based re- construction error to detect anomalies from network flow features. Xuet al.[37] proposed a Log-Cosh Conditional VAE, where the log-cosh loss improves robustness to outliers during reconstruction. These studies show that reconstruction-based VAE models can provide an effective unsupervised or semi- supervised mechanism for detecting abnormal traffic patterns. VAEs have also been extended to specific IDS scenarios. In IoT environments, Khanamet al.[34] incorporated focal loss into a VAE-based model to improve the detection of abnormal traffic under imbalanced conditions. Nguyenet al.[33] pro- posed the Gradient-based Explainable Variational Autoencoder (GEE), which combines anomaly detection with gradient- based explanations to help interpret detected anomalies. For insider threat detection, Pantelidiset al.[32] used autoencoder- based models, including VAEs, to identify abnormal user behavior within an organization. In critical infrastructure, VAE-based methods have been applied to power systems and Advanced Metering Infrastructure (AMI) networks to detect 7 Data Generation EvaluationEffectiveness in Training IDS Models [39], [40] Use Cases Balancing Imbalanced Datasets [41], [42] Simulation of Network Traffic [43], [44] Synthetic Data Creation Enhancing Training Dataset [45], [46] Methods for Generating synthetic data [47], [48] Fig. 3: Variational autoencoder in intrusion detection system data corruption and stealth cyber-attacks [30], [31]. Evaluation studies further show that VAE-based anomaly detection can be improved through model design and loss- function selection. Conditional VAEs have been evaluated for network intrusion detection and shown to be effective when the model is designed to capture class- or condition- specific traffic patterns [28], [29]. Other studies have explored latent-space clustering and inferential autoencoder designs to improve anomaly separation and adapt to changing network conditions [26], [27]. Overall, VAE-based anomaly detection is useful for IDS because it can model normal traffic without requiring extensive labeled attack samples, but its performance depends on the quality of the learned latent representation and the threshold used to distinguish normal and abnormal reconstruction behavior. 2) Data Generation:VAEs can be used to generate syn- thetic IDS data by sampling from the learned latent distribution and decoding the sampled representations into new traffic records. This capability is useful when real network traffic is limited, sensitive, or imbalanced. In IDS, VAE-based data generation is mainly used to enrich training datasets, simulate network traffic, and generate samples for underrepresented attack classes. Renet al.[45] and Linet al.[46] showed that VAE-generated data can improve IDS training by increasing data diversity. Martinet al.[47], [48] further developed VAE- based generative models for intrusion detection and demon- strated their ability to produce synthetic traffic samples for downstream detection tasks. Another application of VAE-based data generation is traffic simulation. In this setting, the generated samples are used to represent different network behaviors, including benign activities and malicious attacks, under controlled experimental conditions. Dinhet al.[43] and Yanget al.[44] used VAE- based models to support IDS training by generating or re- constructing traffic patterns that improve the representation of network behaviors. This is useful for evaluating IDS models when collecting sufficient real-world attack traffic is difficult. VAEs have also been used to address class imbalance by generating synthetic samples for minority attack classes. Class imbalance is common in IDS datasets because benign traffic and frequent attack types often dominate the data, while rare attacks have limited training samples. Chuang and Huang [41] and Yanget al.[42] showed that VAE- Data Augmentation Evaluation Comparison with Other Augmentation Methods [49], [50] Impact on Model Performance [51], [52] ApplicationsImproving Model Robustness [53], [54] Techniques Feature Augmentation [55] Oversampling [44], [56] Fig. 4: Variational autoencoder in intrusion detection system based balancing strategies can improve IDS performance on underrepresented classes. Evaluation studies also indicate that the usefulness of generated data should be measured through downstream IDS performance, such as detection accuracy, recall, and robustness, rather than only by visual or statistical similarity [39], [40]. 3) Data Augmentation:VAE-based data augmentation aims to expand IDS training data by generating additional samples or feature variations from the learned latent distribution. This is different from general data generation because the main objective is not only to create synthetic traffic, but also to improve downstream IDS training. In particular, VAE-based augmentation is useful when the original dataset is limited, imbalanced, or insufficiently diverse. One common strategy is feature-level augmentation, where a VAE learns latent representations of traffic records and generates variations that enrich the original feature space. This strategy can help the IDS model observe broader traffic patterns during training and reduce overfitting to limited samples. For example, Sabeelet al.[55] applied VAE-based augmentation to improve the detection of atypical attack flows, where rare or unusual attack behaviors are difficult to learn from the original data alone. Another important strategy is oversampling, where VAEs generate synthetic samples for underrepresented classes. As noted above, when minority attack classes have too few sam- ples, the detector may become biased toward majority classes and fail to recognize rare attacks. VAE-based oversampling mitigates this issue by increasing the number of minority-class samples while preserving the learned data distribution [44], [56]. Several studies have evaluated the impact of VAE-based augmentation on IDS performance. Compared with traditional augmentation or resampling methods, VAE-based approaches can generate samples that better reflect the underlying feature relationships in network traffic [49], [50]. Empirical results also show that VAE-augmented data can improve detection accuracy, robustness, and class-level performance, especially under imbalanced or limited-data settings [51], [52]. VAE-based augmentation has also been explored for im- proving IDS robustness. By exposing the detector to recon- structed, perturbed, or boundary-related traffic samples, the augmented training data can help the model become less sensitive to small variations in traffic patterns. For example, 8 VAE-based methods have been used in two-stage detection and known/unknown intrusion detection settings to improve the model’s ability to handle abnormal or unseen traffic [53], [54]. Overall, VAE-based data augmentation supports IDS by improving training data diversity, but its effectiveness still depends on the quality of generated samples and their consistency with realistic network behavior. B. GAN in IDS GAN in IDS General applications Tabular data generation [57], [58] Data Imputation/Augmentation [59]–[62] Applications in IDS Adversarial traffic data generation [63]–[65] Network traffic data generation [66] Language model enhanced GAN data augmentation [67] Classical GAN model [68]–[70] GAN related Surveys [71]–[73] Fig. 5: Generative Adversarial Nets (GAN). Generative Adversarial Networks (GANs) were introduced by Goodfellowet al.[68]. A GAN contains two neural networks: a generator and a discriminator. The generator learns to produce synthetic samples, while the discriminator learns to distinguish generated samples from real samples. Through this adversarial training process, the generator gradually im- proves its ability to produce data that resemble the original distribution. Later studies improved the stability and quality of GAN training, including Wasserstein GAN and improved Wasserstein GAN [69], [70]. As summarized in Figure 5, GANs have been applied to IDS mainly for tabular data generation, data augmentation, data imputation, network traffic generation, and adversarial traffic generation. 1) Tabular GAN:Many IDS datasets are represented as tabular data, where each row corresponds to a traffic record and each column corresponds to a feature such as duration, packet count, byte count, protocol type, or flow statistic. Therefore, tabular GAN models are relevant to IDS because they can generate synthetic flow-level records for training or evaluation. Parket al.[57] introduced Table-GAN for synthetic table generation, aiming to reduce the risk of exposing real data during data sharing. Xuet al.[58] proposed CTGAN and TVAE to model complex tabular distributions, including mixed continuous and discrete features. These tabular generation methods provide the methodological basis for applying GANs to flow-based IDS datasets. Conditional GANs have also been used to generate class- specific IDS samples. Liet al.[67] proposed a BERT- enhanced Conditional GAN for multi-class intrusion detection. In this framework, the CGAN generates additional samples for minority attack classes, while BERT is embedded in the discriminator to strengthen the dependency between input features and output labels. The method addresses class im- balance and improves multi-class IDS performance on several datasets, including CSE-CIC-IDS2018, NF-ToN-IoT-V2, and NF-UNSW-NB15-v2. 2) IDS Data Generation:In IDS, GAN-based data gener- ation is mainly used for three purposes. The first purpose is data augmentation, where GANs generate additional samples to enrich the training set and improve downstream detection performance. This is useful when attack samples are limited or when minority attack classes are underrepresented. Park et al.[61] and Huang and Lei [62] applied GAN-based augmentation to address class imbalance in IDS and improve detection performance. Related surveys also show that GANs have been widely studied for imbalance learning and synthetic data generation in broader machine learning settings [74], [75]. The second purpose is data imputation, where GANs esti- mate missing or incomplete feature values while preserving the structure of the original data. Missing values can occur because of packet loss, incomplete collection, preprocessing errors, or unavailable traffic attributes. GAN-based imputation methods have been reviewed in [59], [60], and they are relevant to IDS because incomplete traffic records can reduce the reliability of model training and evaluation. The third purpose is adversarial traffic generation. In this setting, GANs generate malicious traffic records that pre- serve attack functionality while appearing similar to benign or normal traffic. Such generated samples can be used of- fensively to bypass IDS models or defensively to evaluate and improve IDS robustness. De Araujo-Filhoet al.[63], Aldhaheri and Alhuzali [64], and Shuet al.[65] studied GAN- based adversarial generation against IDS. In addition, Zhao et al.[66] investigated GAN-based network traffic generation for improving IDS performance. These studies show that GANs can support IDS model training, but they also introduce security concerns because generated traffic can be used to test or evade detection systems. C. Diffusion in IDS Generative Diffusion Surveys Text Generation Surveys [76], [77] General Surveys [78] Fig. 6: Generative diffusion on surveys. Diffusion models are generative models that learn to gen- erate data through a gradual adding noising and denoising process. Sohl-Dicksteinet al.[79] introduced the diffusion- based generative framework in 2015, where the forward pro- cess progressively adds noise to data and the reverse process 9 learns to recover clean samples from noisy inputs. Later, Denoising Diffusion Probabilistic Models (DDPMs) formal- ized this process as an effective deep generative modeling approach [80]. Although diffusion models were first widely studied in image generation, they have also been extended to text and tabular data generation [76]–[78]. Dhariwal and Nichol [81] further showed that diffusion models can achieve strong image generation performance compared with GAN- based methods. For IDS, diffusion models are relevant because network traffic can be represented in several forms, including tabular flow features, packet sequences, logs, and transformed image- like representations [82], [83]. Therefore, diffusion models can be used for different IDS-related tasks, including adversarial attack generation, adversarial purification, adversarial training, and synthetic data generation. The following subsections sum- marize these applications and highlight their relevance to IDS. 1) Adversarial Attacks:Diffusion models have been studied in cybersecurity partly because they can generate high-quality synthetic samples that resemble real data. Existing surveys and studies have discussed the role of generative models, including diffusion-related methods, in creating adversarial examples and evaluating security vulnerabilities [84]–[87]. In adversarial attack settings, the generated sample is designed to remain close to a valid input while causing a target ML model to make an incorrect prediction. This idea is relevant to IDS because attackers may modify traffic features while preserving malicious behavior, making the attack harder to detect. Some diffusion-based adversarial methods combine the denoising process with gradient-based perturbation strate- gies [88]–[90], such as Projected Gradient Descent (PGD). The purpose is to generate samples that appear realistic but still mislead the target model. Although much of this work has been developed in image domains, the same principle is important for IDS because generated or perturbed traffic can be used to test whether a detector is robust to adaptive attacks. In response to diffusion-generated fake or adversarial sam- ples, Hoodaet al.[91] proposed Disjoint Diffusion Deepfake Detection (D4). D4 is designed to detect fake images gen- erated by diffusion models and to generalize to unseen data distributions and generative techniques. While this work is not specific to IDS, it reflects a broader security problem: as generative models become stronger, detection systems must also be evaluated against generated and previously unseen samples. 2) Adversarial Purification:Adversarial purification (AP) is a defense strategy that uses a generative model to remove or reduce adversarial perturbations before classification. The main idea is to map a potentially perturbed input back toward the clean data distribution so that the downstream classifier receives a less corrupted sample. Early work such as PixelDe- fend used generative modeling for purification [103]. More recent studies have applied diffusion models to adversarial pu- rification because the denoising process can naturally remove small perturbations from input data [94], [97], [105]. Diffusion-based AP has been evaluated in several domains, including image, 3D point cloud, text, and audio data [95], [96], [98]–[102]. Carliniet al.[92] further studied the ro- Adversarial Purification Evaluation [92]–[94] Applications Image domain [95]–[97] 3D image domain [98], [99] Text domain [100], [101] Audio domain [102] Techniques [103]–[105] Fig. 7: Generative diffusion on adversarial purification. bustness guarantees of diffusion-based purification, while other studies examined its limitations and evaluation reliability [93], [104]. For IDS, adversarial purification is a potential direc- tion because network traffic may be intentionally perturbed to evade detection. However, applying AP to IDS requires preserving protocol validity and attack semantics, not only removing statistical noise. Adversarial Training Application [106] Data driven Data size is important [107], [108] Expand dataset size [109]–[112] Adversarial Robustness [113] Fig. 8: Generative diffusion on adversarial training. 3) Adversarial Training:Adversarial training (AT) im- proves model robustness by training the model with adversar- ially perturbed examples. Goodfellowet al.[113] introduced adversarial examples and demonstrated that including such examples during training can improve resistance to adversarial attacks. Later studies showed that robust generalization often requires more training data, because the model must learn stable decision boundaries under both clean and adversarial conditions [107], [108]. This observation motivated the use of external data and generative models to expand training sets for robust learning [109]–[111]. Diffusion models can support adversarial training by gener- ating additional training samples that improve data diversity. Wanget al.[112] used an Elucidating Diffusion Model (EDM) to generate high-quality synthetic image data for adversarial training and showed that generated data can improve ro- bustness without relying only on external real data. Yuet al.[106] proposed the Adversarial Denoising Diffusion Model (ADDM) for unsupervised anomaly detection and showed that it can improve performance compared with DDPM-based 10 anomaly detection methods under reduced sample settings. These studies suggest that diffusion-generated data may be useful for improving IDS robustness, especially when real adversarial or rare attack samples are limited. Data Generation Applications Intrusion detection [114]–[116] Address imbalance data [115] Techniques Image and text [117]–[119] Tabular data [120] Fig. 9: Generative diffusion on tabular data generation. 4) Data Generation:Data generation is the most direct application of diffusion models to IDS [121]. Many IDS datasets are represented as tabular flow-level features, where each record describes a network flow using statistical attributes such as packet counts, byte counts, duration, and protocol in- formation. Therefore, tabular diffusion models are particularly relevant to IDS. TabDDPM [120] extended diffusion modeling to tabular data and provided a basis for generating structured records with both numerical and categorical features. Recent studies have applied diffusion models to intrusion detection and related security tasks [114]–[116]. In these settings, diffusion-generated samples can be used to increase data diversity, supplement limited attack samples, or miti- gate class imbalance. Wanget al.[115] further showed that diffusion-based generation can help address data imbalance by generating additional samples for underrepresented categories. Compared with GAN-based generation, diffusion models may provide more stable training [121], but their usefulness for IDS still depends on whether the generated traffic preserves realistic feature relationships, protocol constraints, and attack semantics. D. LLM in IDS LLM Applications Intrusion detection system [122], [123] Tabular data generation [124]–[127] Explainable IDS IDS alarm explanation [128], [129] Attack classification [130]–[132] Fig. 10: Large Language Model (LLM). Large Language Models (LLMs) are transformer-based models trained to process and generate sequential data. In IDS research, LLMs and related transformer-based language models can be applied when network traffic is represented as logs, packet-byte sequences, flow records converted into textual formats, or structured tabular records. Compared with conventional ML models, LLMs provide two potential ad- vantages for IDS: they can model contextual relationships in sequential traffic representations, and they can generate natural-language explanations for security analysts. However, their use in IDS also introduces practical challenges, including high computation cost, data formatting sensitivity, limited interpretability, and possible hallucination when explanations are generated without sufficient grounding. 1) IDS Applications:LLM-based and transformer-based IDS studies mainly focus on using language-model architec- tures to classify network activities. Liraet al.[122] proposed BERTIDS, a BERT-based model for network intrusion detec- tion. In this method, network logs are converted into tokenized sequences that can be processed by BERT. The model is then fine-tuned to distinguish normal traffic from different attack categories. Their experiments on NSL-KDD reported an accuracy of 98.01% and a false positive rate of 1.48%, showing that transformer-based language models can be adapted to IDS classification tasks. Manocchioet al.[123] presented FlowTransformer, a modu- lar framework for transformer-based Network Intrusion Detec- tion Systems (NIDSs). The framework allows different com- ponents, including input encoding, transformer architecture, classification head, and evaluation dataset, to be replaced and compared. Their evaluation across public flow-based NIDS datasets showed that the classification head has a substantial effect on detection performance. This result indicates that applying transformers to IDS is not only a matter of selecting a large model; the representation of traffic features and the design of the output classifier are also important. Although these studies show the potential of transformer- based models for IDS, several limitations remain. First, fine- tuning large models requires substantial computation and memory resources, which may be impractical for resource- constrained security environments. Second, IDS data are often numerical or mixed-type tabular records, while LLMs are originally designed for textual sequences. Therefore, the per- formance of LLM-based IDS depends strongly on how traffic records are encoded into model-readable inputs. Third, model complexity can make it difficult to interpret why a specific traffic record is classified as malicious, which limits direct use in high-stakes security operations. 2) LLM for Explainable IDS:Another important use of LLMs in IDS is explanation generation. IDS alerts often con- tain technical information such as attack labels, traffic features, source and destination attributes, and model confidence scores. These alerts may be difficult for non-expert users to interpret. LLMs can be used as an explanation layer that converts IDS outputs into natural-language descriptions, summarizes possible causes, and suggests response actions. In this setting, the LLM does not necessarily replace the detector; instead, it helps users understand and act on detection results. Rjoubet al.[128] discussed the role of explainable AI in cybersecurity and highlighted the need for human- understandable explanations in security systems. Juttneret al.[129] proposed ChatIDS, which uses ChatGPT to explain IDS alerts and provide suggestions to non-expert users. This type of approach can improve the usability of IDS outputs, 11 especially when alerts must be interpreted by operators who do not have deep knowledge of network security or ML models. LLMs have also been combined with traditional ML clas- sifiers and XAI tools. Aliet al.[130] introduced HuntGPT, an intrusion detection dashboard that integrates a Random Forest classifier, XAI methods such as SHAP and LIME, and GPT-3.5 Turbo. The classifier detects anomalies, the XAI methods identify important features behind the prediction, and the LLM presents the result in a more understandable form for analysts. Other studies have also explored explainable AI for attack classification and cybersecurity analysis [131], [132]. These works show that LLMs are useful for improving the communication between IDS models and human users, but the generated explanations should be grounded in detector outputs and verified evidence to avoid unsupported conclusions. 3) LLM Tabular Data Generation:In 2024, Kim et al. [124] investigated the effectiveness of using LLM to gener- ate synthetic data that addresses the class imbalance in tabular data. The paper found that using CSV-style prompting (com- pared to sentence-style in GReaT) can significantly improve the ability of LLM to generate accurate and balanced data, enhancing ML performance for minor classes in imbalanced data. Not all data generated by LLMs is equally valuable and useful to downstream model performance; some samples may be harmful. Thus, assessment of the generated data is vital for any generative model. In 2024, Seedat et al. [125] introduced a method called Curated LLM (CLLM), which aims to generate synthetic tabular data in environments where data is scarce (n <100) and to apply a rigorous data curation process to ensure the quality of the generated data. The CLLM first harnesses the prior knowledge embedded in LLMs, using them to generate synthetic datasets based on a small number of real examples. Then, it relies on a curation mechanism that uses metrics like predictive confidence and uncertainty to filter and refine the generated data, improving its utility for downstream ML tasks. The paper used several real-world datasets to demonstrate the superior performance of CLLM over conventional generators (including CTGAN, TVAE, Tab- DDPM, SMOTE, and GReaT) in the low-data regimes. In many domains where data privacy is crucial, synthetic data generated from real datasets can be used to avoid exposing real-world data. However, synthetic data can still contain the original dataset pattern or details. Differential privacy (DP), which introduces randomness in the data generation, is a promising approach to reducing the risk of re-identifying indi- viduals. In 2024, Tran et al. [133] introduced DP-LLMTGen (Differentially Private LLM-based Tabular data Generators), a novel framework designed to generate synthetic tabular data while preserving DP. The framework utilizes a two-stage fine-tuning procedure with a novel loss function specifically designed for tabular data. The first stage focuses on learning the data format using non-sensitive, randomly generated data with original sensitive data. The second stage fine-tunes the LLM with DP mechanisms to ensure that the generated data maintains privacy while accurately capturing the feature distributions and dependencies of the original dataset. Then, synthetic data are generated by sampling from the fine-tuned LLM. The experiment result showed that the proposed DP- LLMTGen framework is able to effectively generate high- fidelity synthetic tabular data while preserving differential privacy. To address the inefficiency and high computational costs associated with using LLMs for tasks involving tabular data, Einy et al. [126] proposed a selective enrichment approach. Their method aims to use LLMs to enrich tabular data to enhance the performance of classical ML models. LLMs are applied only to specific parts of the data that benefit the most from the additional contextual knowledge that LLMs provide. The result demonstrates that this approach can significantly enhance the performance of ML models on tabular data while maintaining cost-effectiveness. However, in 2024, Xu et al. [127] demonstrated that LLMs are generally inadequate for tabular data generation when used directly or even after traditional fine-tuning. They suggest that due to their autoregressive nature, LLMs struggle to model the complex conditional dependencies and mixture distributions that exist in real-world tabular data. Further, feature ordering becomes more important when the dataset grows, and incorrect feature order can significantly degrade the quality of the generated data. The authors proposed a novel approach called Permutation-aided Fine-tuning (PAFT). Although the results show that PAFT can reproduce underlying relationships in generated data, there is still a significant gap between the current capabilities of LLMs and the requirements for generating realistic synthetic tabular data. E. Challenges and Valuable Research Directions Generative AI has been applied to IDS mainly to ad- dress data-related limitations. When labeled attack samples are limited, generative models can create additional samples for training. When datasets contain missing or incomplete records, generative models can support data imputation by estimating missing values from learned feature relationships. When datasets are imbalanced, generative models can generate samples for minority attack classes and reduce the bias of IDS models toward majority classes. Although these applications are useful, the use of generated data in IDS also introduces several challenges. The first challenge is the quality of synthetic data. Gen- erated samples are usually evaluated through their effect on downstream IDS models, such as whether they improve clas- sification accuracy, recall, or robustness. However, synthetic data do not always improve downstream performance. Low- quality samples may introduce noise, distort class boundaries, or cause the detector to overfit artificial patterns. In addition, the performance of generative models is sensitive to design choices and parameter settings, such as latent-space design in VAEs, training stability in GANs, and noise schedules or sam- pling steps in diffusion models [81], [134]–[137]. Therefore, selecting appropriate generative model configurations remains an important problem for IDS applications. The second challenge is the reliability and realism of synthetic network traffic. Common distribution-level metrics, such as Kullback-Leibler (KL) Divergence [138], Jensen- Shannon (JS) Divergence [139], Wasserstein Distance [140], 12 Server ML model R 2 R 1 R 4 R 3 Local dataset D 2 Local dataset D 3 Local dataset D 4 Local dataset D 1 1 1 2 2 Download ML model Upload local dataset 3 3 Training process (a) Centralized machine learning Server ML model R 2 R 1 R 4 R 3 Local dataset D 2 Local dataset D 3 Local dataset D 4 Local dataset D 1 2 2 1 1 Download global model Upload local model 3 3 Aggregation (b) Federated learning Fig. 11: Centralized VS Federated Learning Fr ́ echet Inception Distance (FID) [141], Maximum Mean Dis- crepancy (MMD) [142], Perceptual Path Length (PPL) [143], Energy Distance [144], and Precision and Recall for Distribu- tions [145], can measure similarity between real and generated data distributions. However, distributional similarity alone is not sufficient for IDS. Synthetic traffic should also preserve protocol constraints, temporal dependencies, attack semantics, and network-topology relationships. For example, generated traffic may be statistically similar to real traffic but still invalid in a real testbed if packet sequences violate protocol behavior or if flows do not match the underlying network topology. Future work should develop IDS-specific evaluation methods that assess both statistical fidelity and network-level validity. The third challenge is the limited use of Large Language Models (LLMs) for IDS-specific data generation. Existing LLM-based generation methods have shown potential for text and tabular data generation, including prompt-based synthetic data generation [146]–[148]. However, IDS data often contain numerical features, protocol-dependent relationships, temporal patterns, and topology-aware constraints, which are difficult for general-purpose LLMs to model directly. At the same time, LLMs provide a potential advantage because they can incorporate textual descriptions, domain knowledge, network configurations, and attack procedures during generation. This makes LLMs a promising direction for topology-aware and knowledge-guided IDS data generation. Future research should investigate how to adapt LLMs to network-security domains, how to ground generated traffic in valid network behavior, and how to evaluate whether LLM-generated samples are useful for IDS training and testing. Overall, generative AI can support IDS by improving data availability, diversity, and robustness. However, future studies should move beyond simply generating more samples. More attention is needed on sample quality control, IDS-specific re- alism evaluation, privacy preservation, adversarial misuse, and domain-specific generative models that understand network protocols, attack behaviors, and deployment environments. IV. GENERATIVEAI EMBEDDEDFEDERATEDLEARNING BASEDINTRUSIONDETECTIONSYSTEM FL-based IDS with generative AI LLM embedded FL Diffusion embedded FL [149] GAN embedded FL [64], [150]–[153] VAEs embedded FL [154] Fig. 12: Generative AI with intrusion detection system Federated Learning (FL) is a distributed learning paradigm designed for scenarios where data are generated and stored across many clients. It was originally motivated by appli- cations such as mobile-device intelligence, where user data are distributed across devices and cannot be easily collected at a central server because of privacy and communication constraints [3]. Instead of transferring raw data to the server, FL allows clients to train models locally and share model updates for aggregation. Figure 11 illustrates the difference between centralized Machine Learning (ML) and FL. 13 In centralized ML, as shown in Figure 11a, clients send their local datasets to a central server. The server then combines the collected data, trains an ML model, and distributes the trained model for future prediction. This approach is simple to imple- ment when data can be centrally collected, but it may expose sensitive information and introduce high communication costs when the local datasets are large. In FL, as shown in Figure 11b, training is performed through repeated collaboration between the server and clients. The server first initializes a global model and sends it to participating clients. Each client trains the model using its local dataset and returns the updated model parameters or gradients to the server. The server then aggregates these local updates to obtain a new global model and sends the updated global model back to clients for the next training round. FedAvg is a representative aggregation method that computes a weighted average of local model updates to construct the global model [3]. By avoiding direct raw-data sharing, FL can reduce privacy risks and raw-data transmission costs, although communication overhead and potential information leakage from model updates remain important concerns. FL is particularly relevant to IDS because network traffic is naturally distributed across routers, edge devices, organi- zations, and geographic regions. In a traditional centralized IDS training pipeline, clients or network devices send traffic records to a central server, and the server uses the aggregated data to train a detection model. This process can be expensive for high-volume traffic and may expose sensitive information, such as user behavior, service configurations, or organization- specific security patterns. The problem becomes more sig- nificant for high-rate attacks [155], such as denial-of-service (DoS) and distributed denial-of-service (DDoS) attacks, which can generate a large number of traffic records in a short period. FL-based IDS addresses this limitation by allowing each client to train locally while sharing only model updates with the server. However, FL-based IDS also introduces challenges that are different from centralized IDS. Client data are often non-independent and identically distributed (non-IID) [156] because different clients may observe different traffic volumes, device types, services, and attack categories. Some clients may have limited attack samples, while others may have highly imbalanced traffic distributions. In addition, clients may have different computation and communication capabilities, and the FL process may be vulnerable to poisoning attacks or unreliable updates. These challenges motivate the integra- tion of generative AI with FL-based IDS. Generative models can potentially augment local data, mitigate class imbalance, support privacy-preserving synthetic data generation, improve robustness, and reduce the effect of heterogeneous client distributions. This section reviews generative AI-embedded FL-based IDS according to the same generative model families discussed in Section I. Section IV-A discusses VAE-embedded FL-IDS, where autoencoder-based models can support representation learning, anomaly detection, and communication reduction. Section IV-B reviews GAN-embedded FL-IDS, with a fo- cus on data augmentation, adversarial traffic generation, and class-imbalance mitigation. Section IV-C discusses diffusion- embedded FL-IDS and its potential for synthetic data genera- tion under distributed settings. Finally, Section IV-D discusses the emerging role of LLMs in FL-based IDS, including their potential use for explanation, knowledge-guided generation, and network-security analysis. A. Autoencoder-embedded FL-IDS Autoencoder-based models, including Autoencoders (AEs) and Variational Autoencoders (VAEs), have been used in FL-based IDS mainly for representation learning, anomaly detection, privacy preservation, and communication reduction. An AE learns to compress input data into a lower-dimensional latent representation and reconstruct the original input from that representation. This design is useful for IDS because net- work traffic often contains high-dimensional flow features, and the learned latent representation can preserve important traffic patterns while reducing feature size. A VAE extends the AE by modeling the latent representation as a probability distribution, which enables sampling and synthetic data generation from the learned distribution. In FL-based IDS, autoencoder-based models are useful because clients can learn compact representations of local traffic without directly sharing raw network data. Instead of transmitting full datasets, clients may share model updates or compressed latent representations with the server. This can reduce communication cost and limit direct exposure of sensitive traffic records. In addition, reconstruction error from an AE or VAE can be used for anomaly detection, where traffic samples with large reconstruction errors are treated as potential intrusions [54], [157]–[159]. Tayeenet al.[160] proposed CAFNET, a compressed autoencoder-based federated framework for network anomaly detection. CAFNET reduces communication overhead by transmitting compact representations between clients and the server rather than transferring raw data or large model com- ponents. The reported results show that CAFNET maintains detection performance while reducing communication cost by up to 95%. NRet al.[154] also used VAE-based federated learning for intrusion detection in Industrial IoT environments, aiming to improve privacy protection and reduce communica- tion overhead. Overall, AE- and VAE-embedded FL-IDS methods are suitable for settings where network traffic is high-dimensional, privacy-sensitive, and distributed across multiple clients. How- ever, these methods still face several challenges. The com- pressed latent representation must preserve enough informa- tion for accurate intrusion detection, while avoiding unnec- essary leakage of sensitive traffic patterns. In addition, when client data are non-IID, the latent spaces learned by different clients may not be fully aligned, which can affect aggregation and global anomaly detection performance. B. GAN-embedded FL-IDS GANs can be integrated with FL-based IDS from two main perspectives. The first perspective treats GAN-generated traffic as a security threat. As discussed in Section I-B2, GANs can 14 generate adversarial traffic records that resemble real traffic and are designed to bypass IDS models. For example, Linet al.[152] introduced IDSGAN for generating adversarial attack traffic against IDS. Aldhaheri and Alhuzali [64] proposed SGAN-IDS, and Zhanget al.[153] studied poisoning attacks against FL-based network intrusion detection. These studies show that generated malicious traffic can be difficult to detect because it is optimized to remain close to legitimate traffic patterns while misleading the detection model. In FL-based IDS, this problem becomes more complex because the server does not directly access clients’ raw traffic data. The server mainly observes model updates, which makes it harder to inspect whether local training data contain GAN- generated adversarial samples or poisoned records. Therefore, detecting and defending against GAN-based adversarial traffic remains an important challenge for FL-based IDS. Vyet al.[150] studied poisoning attacks and defense mechanisms in an FL-based IDS framework for Industrial IoT networks, showing that adversarial manipulation must be considered when deploying IDS models in distributed environments. The second perspective uses GANs as a defensive tool for data augmentation. In FL-based IDS, clients often hold non- IID and imbalanced local datasets because different networks observe different services, devices, traffic volumes, and attack categories. A local GAN can generate additional samples for minority classes and improve the local training distribution without requiring clients to share raw traffic data. Tabassum et al.[151] proposed FEDGAN-IDS, where GAN-based aug- mentation is used with FL to address data imbalance while preserving privacy. Despite this potential, GAN-embedded FL-IDS faces com- munication and training challenges. A GAN usually contains at least two model components, a generator and a discriminator, which increases training complexity and may introduce addi- tional communication cost if the GAN parameters are shared across clients and the server. Model compression and scaling studies, such as [161], [162], show that model size, input size, and performance must be balanced carefully. For FL-based IDS, this means that GANs should be designed or deployed in a way that improves local data quality without offsetting the privacy and communication advantages of FL. C. Diffusion-embedded FL-IDS Diffusion models are an emerging direction for generative AI-embedded FL-based IDS. Compared with GANs, diffusion models often provide more stable training because they do not rely on adversarial optimization between a generator and a discriminator. This property is useful in FL, where clients may have heterogeneous and imbalanced local data. However, diffusion models can also require substantial sampling com- putation, so their communication and computation costs must be evaluated carefully in distributed IDS settings. Jothiraj and Mashhadi [149] introduced Phoenix, a federated generative diffusion framework that uses diffusion models to improve training data diversity across clients. Their study compared diffusion-based generation with GAN-based gener- ation in an FL setting and showed that diffusion models can generate high-quality samples while reducing communication cost. Although Phoenix is not limited to IDS, its design is relevant to FL-based IDS because local clients often have limited, imbalanced, and non-IID traffic data. For FL-based IDS, diffusion models can potentially support local data augmentation, minority-class sample generation, and privacy-preserving synthetic traffic generation. Instead of shar- ing raw traffic data, clients may use diffusion models to enrich local training data or share generative knowledge with the server. This direction is especially useful when attack samples are rare or unevenly distributed across clients. Nevertheless, diffusion-embedded FL-IDS remains underexplored. Future studies need to examine whether diffusion-generated traffic preserves realistic network behavior, whether the generation process is efficient enough for edge or IoT clients, and how diffusion models can be integrated with FL without increasing communication overhead. D. LLM-embedded FL-IDS LLM-embedded FL-IDS remains an emerging direction with limited direct studies. In this context, an LLM can be integrated into an FL-based IDS pipeline in several ways, including traffic classification, anomaly detection, data aug- mentation, data imputation, alert explanation, and security knowledge extraction. These capabilities are relevant to FL- based IDS because clients often have limited, imbalanced, and heterogeneous local traffic data. For example, LLM-based data augmentation or imputation may help enrich local datasets without requiring clients to share raw traffic records. However, applying LLMs directly in FL-based IDS is chal- lenging. The main limitation is the high communication and computation cost of training or fine-tuning large models across distributed clients. Unlike smaller IDS models, LLMs contain a large number of parameters, making full-model transmission impractical for many edge, IoT, or organizational clients. In addition, IDS data are often represented as numerical flow features, packet sequences, or structured logs, while general- purpose LLMs are primarily trained on natural-language text. Therefore, effective input representation and domain adapta- tion are necessary before LLMs can be reliably used for IDS tasks. A practical direction is to avoid transmitting full LLMs in the FL process. Instead, future studies may explore parameter- efficient tuning, adapter-based learning, knowledge distillation, or server-side LLM assistance. In these settings, clients may train lightweight local IDS models, while LLMs provide aux- iliary functions such as generating domain-informed synthetic samples, explaining alerts, summarizing attack behaviors, or supporting security analysts. Overall, LLM-embedded FL- IDS has potential, but its feasibility depends on reducing communication cost, grounding LLM outputs in valid network behavior, and adapting LLMs to network-security-specific data formats. E. Challenges and Valuable Research Directions Generative AI-embedded FL-based IDS introduces both opportunities and risks. On the defensive side, generative 15 models can support local data augmentation, mitigate class imbalance, improve robustness, and reduce the need to share raw traffic data. On the offensive side, the same generative models can be used to create synthetic malicious traffic that resembles benign traffic or bypasses IDS models. This dual- use nature makes generative AI important for both improving FL-based IDS and evaluating its vulnerability under adaptive attacks. The first research challenge is defending FL-based IDS against generative adversarial traffic. GANs and diffusion models can generate traffic samples that are statistically close to real traffic but intentionally optimized to mislead a detector. In FL-based IDS, this threat is harder to identify because the server usually receives model updates rather than raw local traffic. As a result, adversarial or poisoned local data may influence the global model through aggregation without being directly inspected. Future work should study how to detect generative adversarial traffic in distributed settings, how to distinguish malicious client updates from benign non-IID updates, and how to design robust aggregation methods for FL-based IDS. The second research direction is using generative AI to address non-independent and identically distributed (non-IID) client data. In FL-based IDS, each client may observe different devices, services, traffic volumes, and attack categories. This heterogeneity can reduce the quality of the global model because local updates are optimized on different data dis- tributions. Generative models can reduce this problem by augmenting local datasets, generating minority-class samples, or improving local data diversity before model training. Exist- ing studies have explored VAE-, GAN-, and diffusion-based methods for handling non-IID data in FL [163]–[165]. For IDS, this direction is especially relevant because rare attacks may appear only on a small subset of clients. The third challenge is communication efficiency. Although generative models can improve local training data, transmitting large generative models between the server and clients may increase communication cost. This problem is more significant for GANs and LLMs because they may contain large model components or require expensive fine-tuning. In contrast, some VAE- and diffusion-based FL methods can be designed to reduce communication by sharing compact representations, selected parameters, or generated knowledge instead of full datasets or full models. Recent studies on communication- efficient federated diffusion learning show that diffusion-based strategies can reduce communication cost while maintaining model performance [166], [167]. Future FL-based IDS studies should evaluate not only detection accuracy, but also commu- nication cost, client computation cost, and scalability. The fourth research direction is realistic federated IDS data generation and benchmarking. As discussed in Section I-E, realistic FL-based IDS datasets remain limited. Most FL-based IDS studies use centrally collected datasets and partition them artificially across clients. Although this strategy is convenient, it may not reflect real client-level heterogeneity, network topol- ogy, temporal changes, or organization-specific attack patterns. FLNET [19] provides an important step toward FL-oriented IDS evaluation, but more datasets are needed to represent diverse federated deployment scenarios. Generative AI may help create topology-aware and client-specific traffic data, but the generated data must be evaluated for both statistical similarity and network-level validity. The fifth research direction is LLM-assisted FL-based IDS. To the best of our knowledge, direct studies on LLM- embedded FL-based IDS are still limited. However, LLMs may support FL-based IDS through domain-informed data generation, alert explanation, attack-behavior summarization, traffic-log analysis, and response recommendation. Compared with VAEs, GANs, and diffusion models, LLMs may be better suited for incorporating textual domain knowledge, such as protocol descriptions, network configurations, and attack pro- cedures. However, most pre-trained LLMs are general-purpose models and are not optimized for network-security data. Fu- ture studies should investigate network-domain-specific LLMs, parameter-efficient adaptation, and methods for grounding LLM outputs in valid traffic behavior and verified security evidence. Overall, generative AI-embedded FL-based IDS should be evaluated from multiple perspectives, including detection per- formance, robustness to adversarial generation, privacy pro- tection, communication efficiency, and realism of generated traffic. Future research should move beyond using generative models only as data generators and study how they can be safely integrated into distributed IDS training and deployment. V. CONCLUSION This survey examined the intersection of generative AI, Federated Learning (FL), and Intrusion Detection Systems (IDSs). We reviewed representative IDS research direc- tions—adversarial machine learning, anomaly-based detec- tion, IoT-oriented IDS, explainable IDS, and benchmark datasets—organized generative AI applications in IDS by four model families (VAEs, GANs, diffusion models, and LLMs) and task objectives, and analyzed how these models are being integrated into FL-based IDS. We now return to the four research questions posed in Section I-A and summarize what the reviewed literature indicates. RQ1 (How are generative models used to improve IDS?) The reviewed studies show that generative models address IDS along two complementary axes: model family and task objective. VAEs are used primarily for reconstruction-based anomaly detection, latent representation learning, data gen- eration, and augmentation. GANs are most often applied to tabular flow generation, class-imbalance mitigation, data impu- tation, and, on the offensive side, adversarial traffic generation. Diffusion models are an emerging alternative for synthetic data generation, adversarial purification, and robustness improve- ment, and tend to offer more stable training than GANs at higher sampling cost. LLMs introduce capabilities the other families lack, including traffic-log analysis, natural-language alert explanation, and knowledge-guided tabular generation. These families also differ in their assumptions: VAEs and GANs operate on fixed feature representations, diffusion mod- els trade computation for stability, and LLMs require traffic to be encoded into model-readable formats. Across all families, 16 the dominant motivation is the same: compensating for limited, imbalanced, incomplete, or non-shareable IDS data. RQ2 (How reliable is synthetic network traffic for IDS training and evaluation?) The literature indicates that statis- tical fidelity is necessary but not sufficient. Distribution-level metrics such as KL and JS divergence, Wasserstein distance, FID, and MMD can quantify similarity to real data, but high similarity does not guarantee that generated traffic preserves protocol constraints, temporal dependencies, attack semantics, or network-topology relationships. Several studies report that synthetic data improves some downstream detectors while degrading others, and that low-quality samples can distort decision boundaries. We therefore conclude that synthetic IDS traffic remains only conditionally reliable, and that IDS- specific evaluation—measuring both statistical realism and network-level validity—is still an open requirement rather than a solved problem. RQ3 (How can generative AI support privacy-preserving and communication-efficient FL-based IDS?) The reviewed FL studies show that generative models can augment lo- cal datasets, generate minority-class samples, reduce non-IID skew, and produce privacy-preserving synthetic data without sharing raw traffic. Autoencoder- and VAE-based methods ad- ditionally support communication reduction through compact latent representations, and diffusion-based federated methods have shown that high-quality generation can be compatible with reduced communication cost. At the same time, this ca- pability is dual-use: GAN- and diffusion-generated traffic can poison local updates or evade detection, and because the server observes only model updates, such manipulation is harder to detect in FL than in centralized settings. The integration of generative AI into FL-based IDS is thus promising but still early, supported by relatively few primary studies and an open need for robust aggregation and communication-efficient generative sharing. RQ4 (What datasets and benchmarks are available?) We surveyed widely used IDS datasets (Table I), spanning tradi- tional network intrusion detection, DDoS, IoT, and in-vehicle settings. Most are centrally collected and were not designed to capture client-level heterogeneity, network topology, temporal drift, or organization-specific attack patterns. FLNET2023 is, to our knowledge, the principal dataset that explicitly supports federated evaluation. We therefore conclude that current benchmarks are adequate for centralized generative IDS research but insufficient for federated and topology-aware evaluation, which is one of the most concrete gaps the field faces. Taken together, these answers point to a small set of priorities for future work. First, more realistic and explicitly federated IDS benchmarks are needed, since artificially par- titioned centralized datasets do not reflect real deployment. Second, synthetic traffic generation should become protocol- and topology-aware, and should be evaluated for network- level validity rather than statistical similarity alone. Third, communication-efficient generative FL methods are required so that the cost of sharing generative models does not offset the privacy and efficiency benefits of FL. Fourth, network- domain-specific LLMs, adapted through parameter-efficient tuning and grounded in verified traffic behavior, deserve fo- cused study for IDS data generation, alert explanation, and response recommendation. Finally, because generative models are inherently dual-use, FL-based IDS should be evaluated not only on detection accuracy but also on robustness to adver- sarial generation, privacy leakage, and communication cost. Building reliable generative AI-enabled IDS will ultimately require combining accurate detection, realistic data generation, privacy preservation, communication efficiency, and security- aware evaluation within a single framework. ACKNOWLEDGMENT This work was supported by the DEVCOM Analysis Center under Cooperative Agreement Nos. W911QX23D0009 and W911NF2220001, and in part by NSF awards 2148358, 2417062, and 1914635. REFERENCES [1] James P Anderson. Computer security threat monitoring and surveil- lance.Technical Report, James P. Anderson Company, 1980. [2] Dorothy E. Denning.An intrusion-detection model.IEEE Trans. Software Eng., 13(2):222–232, 1987. [3] Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Ag ̈ uera y Arcas.Communication-efficient learning of deep networks from decentralized data. In Aarti Singh and Xiaojin (Jerry) Zhu, editors,Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, AISTATS 2017, 20-22 April 2017, Fort Lauderdale, FL, USA, volume 54 ofProceedings of Machine Learning Research, pages 1273–1282. PMLR, 2017. [4] Houda Jmila and Mohamed Ibn Khedher. Adversarial machine learn- ing for network intrusion detection: A comparative study.Comput. Networks, 214:109073, 2022. [5] Mahbod Tavallaee, Ebrahim Bagheri, Wei Lu, and Ali A Ghorbani. A detailed analysis of the kdd cup 99 data set. In2009 IEEE symposium on computational intelligence for security and defense applications, pages 1–6. Ieee, 2009. [6] Nour Moustafa and Jill Slay. Unsw-nb15: a comprehensive data set for network intrusion detection systems (unsw-nb15 network data set). In2015 military communications and information systems conference (MilCIS), pages 1–6. IEEE, 2015. [7] Afnan Alotaibi and Murad A. Rassam. Adversarial machine learning attacks against intrusion detection systems: A survey on strategies and defense.Future Internet, 15(2), 2023. [8] Zhen Yang, Xiaodong Liu, Tong Li, Di Wu, Jinjiang Wang, Yunwei Zhao, and Han Han.A systematic literature review of methods and datasets for anomaly-based network intrusion detection.Comput. Secur., 116:102675, 2022. [9] Suzan Hajj, Rayane El Sibai, Jacques Bou Abdo, Jacques Demerjian, Abdallah Makhoul, and Christophe Guyeux. Anomaly-based intrusion detection systems: The requirements, methods, measurements, and datasets.Transactions on Emerging Telecommunications Technologies, 32(4):e4240, 2021. [10] SVN Santhosh Kumar, M Selvi, and A Kannan. A comprehensive survey on machine learning-based intrusion detection systems for secure communication in internet of things.Computational Intelligence and Neuroscience: CIN, 2023, 2023. [11] PLS Jayalaxmi, Rahul Saha, Gulshan Kumar, Mauro Conti, and Tai- Hoon Kim. Machine and deep learning solutions for intrusion detection and prevention in iots: A survey.IEEE Access, 10:121173–121192, 2022. [12] Nour Moustafa, Nickolaos Koroniotis, Marwa Keshk, Albert Y. Zomaya, and Zahir Tari. Explainable intrusion detection for cyber defences in the internet of things: Opportunities and solutions.IEEE Communications Surveys & Tutorials, 25(3):1775–1807, 2023. [13] Subash Neupane, Jesse Ables, William Anderson, Sudip Mittal, Shahram Rahimi, Ioana Banicescu, and Maria Seale.Explainable intrusion detection systems (X-IDS): A survey of current methods, challenges, and opportunities.IEEE Access, 10:112392–112415, 2022. [14] David Gunning and David Aha. Darpa’s explainable artificial intelli- gence (xai) program.AI magazine, 40(2):44–58, 2019. 17 [15] Abhishek Divekar, Meet Parekh, Vaibhav Savla, Rudra Mishra, and Mahesh Shirole. Benchmarking datasets for anomaly-based network intrusion detection: Kdd cup 99 alternatives.In2018 IEEE 3rd International Conference on Computing, Communication and Security (ICCCS), pages 1–8. IEEE, 2018. [16] Iman Sharafaldin, Arash Habibi Lashkari, and Ali A Ghorbani. Toward generating a new intrusion detection dataset and intrusion traffic characterization.ICISSp, 1:108–116, 2018. [17] Iman Sharafaldin, Arash Habibi Lashkari, Saqib Hakak, and Ali A Ghorbani. Developing realistic distributed denial of service (ddos) at- tack dataset and taxonomy. In2019 International Carnahan Conference on Security Technology (ICCST), pages 1–8. IEEE, 2019. [18] Hyunjae Kang, Byung Il Kwak, Young Hun Lee, Haneol Lee, Hwejae Lee, and Huy Kang Kim. Car hacking: Attack & defense challenge 2020 dataset, 2021. [19] Pratyay Kumar, Jiefei Liu, Abu Saleh Md Tayeen, Satyajayant Misra, Huiping Cao, Jayashree Harikumar, and Oscar Perez.Flnet2023: Realistic network intrusion detection dataset for federated learning. InMILCOM 2023-2023 IEEE Military Communications Conference (MILCOM), pages 345–350. IEEE, 2023. [20] Euclides Carlos Pinto Neto, Sajjad Dadkhah, Raphael Ferreira, Alireza Zohourian, Rongxing Lu, and Ali A Ghorbani. Ciciot2023: A real- time dataset and benchmark for large-scale attacks in iot environment. Sensors, 23(13):5941, 2023. [21] Euclides Carlos Pinto Neto, Hamideh Taslimasa, Sajjad Dadkhah, Shahrear Iqbal, Pulei Xiong, Taufiq Rahman, and Ali A Ghorbani. Ci- ciov2024: Advancing realistic ids approaches against dos and spoofing attack in iov can bus.Internet of Things, 26:101209, 2024. [22] Seonghoon Jeong, Sangho Lee, Hwejae Lee, and Huy Kang Kim. X-canids: Signal-aware explainable intrusion detection system for controller area network-based in-vehicle network.IEEE Transactions on Vehicular Technology, 73(3):3230–3246, 2024. [23] Markus Ring, Sarah Wunderlich, Deniz Scheuring, Dieter Landes, and Andreas Hotho. A survey of network-based intrusion detection data sets.Computers & Security, 86:147–167, 2019. [24] Ansam Khraisat, Iqbal Gondal, Peter Vamplew, and Joarder Kamruz- zaman. Survey of intrusion detection systems: techniques, datasets and challenges.Cybersecurity, 2(1):1–22, 2019. [25] Diederik P. Kingma and Max Welling.Auto-encoding variational bayes.In Yoshua Bengio and Yann LeCun, editors,2nd Interna- tional Conference on Learning Representations, ICLR 2014, Banff, AB, Canada, April 14-16, 2014, Conference Track Proceedings, 2014. [26] Van Quan Nguyen, Viet Hung Nguyen, Tuan Hao Hoang, and Nathan Shone. A novel deep clustering variational auto-encoder for anomaly- based network intrusion detection. In Nguyen Ba Hung, Le Sy Vinh, Nguyen Le Minh, Ha Minh Hoang, Bui Thu Lam, Phan Xuan Hieu, Nguyen Trung Lap, Vo Dinh Hieu, and Van Huu Quang Nhat, editors, 14th International Conference on Knowledge and Systems Engineering, KSE 2022, Nha Trang, Vietnam, October 19-21, 2022, pages 1–7. IEEE, 2022. [27] Abdul Hannan, Christian Gruhl, and Bernhard Sick. Anomaly based resilient network intrusion detection using inferential autoencoders. In IEEE International Conference on Cyber Security and Resilience, CSR 2021, Rhodes, Greece, July 26-28, 2021, pages 1–7. IEEE, 2021. [28] Haocheng Xie, Aiping Li, Rong Jiang, Yan Jia, Li Huang, and Weihong Han. Intrusion detection results analysis based on variational auto- encoder. InFourth IEEE International Conference on Data Science in Cyberspace, DSC 2019, Hangzhou, China, June 23-25, 2019, pages 516–521. IEEE, 2019. [29] Shuhana Azmin and A. B. M. Alim Al Islam. Network intrusion detection system based on conditional variational laplace autoencoder. In7th NSysS 2020: 7th International Conference on Networking, Systems and Security, Dhaka, Bangladesh, 22-24 December, 2020, pages 82–88. ACM, 2020. [30] Ruicheng Zheng and Jie Gu. Anomaly detection for power system forecasting under data corruption based on variational auto-encoder. In8th Renewable Power Generation Conference (RPG 2019), pages 1–6. IET, 2019. [31] Abdulrahman Takiddin, Muhammad Ismail, Usman Zafar, and Erchin Serpedin. Variational auto-encoder-based detection of electricity stealth cyber-attacks in AMI networks. In28th European Signal Processing Conference, EUSIPCO 2020, Amsterdam, Netherlands, January 18-21, 2021, pages 1590–1594. IEEE, 2020. [32] Efthimios Pantelidis, Gueltoum Bendiab, Stavros Shiaeles, and Nicholas Kolokotronis. Insider threat detection using deep autoencoder and variational autoencoder neural networks. InIEEE International Conference on Cyber Security and Resilience, CSR 2021, Rhodes, Greece, July 26-28, 2021, pages 129–134. IEEE, 2021. [33] Quoc Phong Nguyen, Kar Wai Lim, Dinil Mon Divakaran, Kian Hsiang Low, and Mun Choon Chan.GEE: A gradient-based explainable variational autoencoder for network anomaly detection. In7th IEEE Conference on Communications and Network Security, CNS 2019, Washington, DC, USA, June 10-12, 2019, pages 91–99. IEEE, 2019. [34] Shapla Khanam, Ismail Ahmedy, Mohd Yamani Idna Bin Idris, and Mohamed Hisham Jaward. Towards an effective intrusion detection model using focal loss variational autoencoder for internet of things (iot).Sensors, 22(15):5822, 2022. [35] Hyunseung Choi, Mintae Kim, Gyubok Lee, and Wooju Kim. Un- supervised learning approach for network intrusion detection system using autoencoders.J. Supercomput., 75(9):5597–5621, 2019. [36] Genki Osada, Kazumasa Omote, and Takashi Nishide.Network intrusion detection based on semi-supervised variational auto-encoder. In Simon N. Foley, Dieter Gollmann, and Einar Snekkenes, editors, Computer Security - ESORICS 2017 - 22nd European Symposium on Research in Computer Security, Oslo, Norway, September 11-15, 2017, Proceedings, Part I, volume 10493 ofLecture Notes in Computer Science, pages 344–361. Springer, 2017. [37] Xing Xu, Jie Li, Yang Yang, and Fumin Shen.Toward effective intrusion detection using log-cosh conditional variational autoencoder. IEEE Internet Things J., 8(8):6187–6196, 2021. [38] Sultan Zavrak and Murat Iskefiyeli. Anomaly-based intrusion detection from network flow features using variational autoencoder.IEEE Access, 8:108346–108358, 2020. [39] Laze Gjorgiev and Sonja Gievska. Time series anomaly detection with variational autoencoder using mahalanobis distance. In Vesna Dimitrova and Ivica Dimitrovski, editors,ICT Innovations 2020. Ma- chine Learning and Applications - 12th International Conference, ICT Innovations 2020, Skopje, North Macedonia, September 24-26, 2020, Proceedings, volume 1316 ofCommunications in Computer and Information Science, pages 42–55. Springer, 2020. [40] Dibyajyoti Guha, Rajdeep Chatterjee, and Biplab Sikdar. Anomaly detection using lstm-based variational autoencoder in unsupervised data in power grid.IEEE Syst. J., 17(3):4313–4323, 2023. [41] Po-Jen Chuang and Pang-Yu Huang. B-VAE: a new dataset balancing approach using batched variational autoencoders to enhance network intrusion detection.J. Supercomput., 79(12):13262–13286, 2023. [42] Yanqing Yang, Kangfeng Zheng, Bin Wu, Yixian Yang, and Xiujuan Wang. Network intrusion detection based on supervised adversarial variational auto-encoder with regularization.IEEE access, 8:42169– 42184, 2020. [43] Phai Vu Dinh, Quang Uy Nguyen, Dinh Thai Hoang, Diep N. Nguyen, Son Pham Bao, and Eryk Dutkiewicz. Constrained twin variational auto-encoder for intrusion detection in iot systems.IEEE Internet Things J., 11(8):14789–14803, 2024. [44] Yanqing Yang, Kangfeng Zheng, Chunhua Wu, and Yixian Yang. Improving the classification effectiveness of intrusion detection by using improved conditional variational autoencoder and deep neural network.Sensors, 19(11):2528, 2019. [45] Yi Ren, Kanghui Feng, Fei Hu, Liangyin Chen, and Yanru Chen. A lightweight unsupervised intrusion detection model based on variational auto-encoder.Sensors, 23(20):8407, 2023. [46] Ying-Dar Lin, Zi-Qiang Liu, Ren-Hung Hwang, Van-Linh Nguyen, Po- Ching Lin, and Yuan-Cheng Lai. Machine learning with variational autoencoder for imbalanced datasets in intrusion detection.IEEE Access, 10:15247–15260, 2022. [47] Manuel L ́ opez Mart ́ ın, Bel ́ en Carro, and Antonio S ́ anchez-Esguevillas. Variational data generative model for intrusion detection.Knowl. Inf. Syst., 60(1):569–590, 2019. [48] Manuel L ́ opez Mart ́ ın, Bel ́ en Carro, Antonio S ́ anchez-Esguevillas, and Jaime Lloret. Conditional variational autoencoder for prediction and feature recovery applied to intrusion detection in iot.Sensors, 17(9):1967, 2017. [49] Kh Shaikh Ahmed and Li Yue. Improving intrusion detection system using improved variational autoencoder. In2023 8th International Conference on Intelligent Computing and Signal Processing (ICSP), pages 215–219, 2023. [50] Po-Jen Chuang and Dong-Ye Wu. Applying deep learning to balancing network intrusion detection datasets.In11th IEEE International Conference on Advanced Infocomm Technology, ICAIT 2019, Jinan, China, October 18-20, 2019, pages 213–217. IEEE, 2019. [51] Vipasha Thakur, Dhandapani Vaithiyanathan, Preeti Verma, and Baljit Kaur. A comparative evaluation of intrusion detection systems: Icvae vs. cvae. In2024 International Conference on Advances in Modern Age 18 Technologies for Health and Engineering Science (AMATHE), pages 1–6, 2024. [52] Chang Liu, Ruslan Antypenko, Iryna Sushko, and Oksana Za- kharchenko.Intrusion detection system after data augmentation schemes based on the VAE and CVAE.IEEE Trans. Reliab., 71(2):1000–1010, 2022. [53] Ryo Yaegashi, Erina Takeshita, and Yu Nakayama. Two-stage ddos mitigation with variational auto-encoder and cyclic queuing. InIEEE International Conference on Communications, ICC 2022, Seoul, Korea, May 16-20, 2022, pages 5421–5426. IEEE, 2022. [54] Jian Yang, Xiang Chen, Shuangwu Chen, Xiaofeng Jiang, and Xi- aobin Tan. Conditional variational auto-encoder and extreme value theory aided two-stage learning approach for intelligent fine-grained known/unknown intrusion detection.IEEE Trans. Inf. Forensics Secur., 16:3538–3553, 2021. [55] Ulya Sabeel, Shahram Shah-Heydari, Khalid Elgazzar, and Khalil El- Khatib. CVAE-AN: atypical attack flow detection using incremental adversarial learning. InIEEE Global Communications Conference, GLOBECOM 2021, Madrid, Spain, December 7-11, 2021, pages 1– 6. IEEE, 2021. [56] Mahmoud Mohamed. Comparative evaluation of vaes, vae-gans and aaes for anomaly detection in network intrusion data.EMITTER International Journal of Engineering Technology, 11(2):160–173, 2023. [57] Noseong Park, Mahmoud Mohammadi, Kshitij Gorde, Sushil Jajodia, Hongkyu Park, and Youngmin Kim. Data synthesis based on generative adversarial networks.Proc. VLDB Endow., 11(10):1071–1083, 2018. [58] Lei Xu, Maria Skoularidou, Alfredo Cuesta-Infante, and Kalyan Veera- machaneni. Modeling tabular data using conditional gan. InAdvances in Neural Information Processing Systems, 2019. [59] Jaeyoon Kim, Donghyun Tae, and Junhee Seok. A survey of miss- ing data imputation using generative adversarial networks. In2020 International Conference on Artificial Intelligence in Information and Communication, ICAIIC 2020, Fukuoka, Japan, February 19-21, 2020, pages 454–456. IEEE, 2020. [60] Reza Shahbazian and Sergio Greco. Generative adversarial networks assist missing data imputation: A comprehensive survey and evaluation. IEEE Access, 11:88908–88928, 2023. [61] Cheolhee Park, Jonghoon Lee, Youngsoo Kim, Jong-Geun Park, Hyun- jin Kim, and Dowon Hong. An enhanced ai-based network intrusion detection system using generative adversarial networks.IEEE Internet of Things Journal, 10(3):2330–2345, 2022. [62] Shuokang Huang and Kai Lei. Igan-ids: An imbalanced generative adversarial network towards intrusion detection system in ad-hoc networks.Ad Hoc Networks, 105:102177, 2020. [63] Paulo Freitas de Araujo-Filho, Mohamed Naili, Georges Kaddoum, Emmanuel Thepie Fapi, and Zhongwen Zhu. Unsupervised gan-based intrusion detection system using temporal convolutional networks and self-attention.IEEE Transactions on Network and Service Manage- ment, 2023. [64] Sahar Aldhaheri and Abeer Alhuzali. SGAN-IDS: self-attention-based generative adversarial network against intrusion detection systems. Sensors, 23(18):7796, 2023. [65] Dule Shu, Nandi O Leslie, Charles A Kamhoua, and Conrad S Tucker. Generative adversarial attacks against intrusion detection systems using active learning. InProceedings of the 2nd ACM workshop on wireless security and machine learning, pages 1–6, 2020. [66] Xinxing Zhao, Kar Wai Fok, and Vrizlynn L Thing. Enhancing network intrusion detection performance using generative adversarial networks.arXiv preprint arXiv:2404.07464, 2024. [67] Fang Li, Hang Shen, Jieai Mai, Tianjing Wang, Yuanfei Dai, and Xiaodong Miao.Pre-trained language model-enhanced conditional generative adversarial networks for intrusion detection.Peer Peer Netw. Appl., 17(1):227–245, 2024. [68] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets.Advances in neural information processing systems, 27, 2014. [69] Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville.Improved training of wasserstein gans. Advances in neural information processing systems, 30, 2017. [70] Mart ́ ın Arjovsky, Soumith Chintala, and L ́ eon Bottou. Wasserstein GAN.CoRR, abs/1701.07875, 2017. [71] Amina Kammoun, Rim Slama, Hedi Tabia, Tarek Ouni, and Mohamed Abid. Generative adversarial networks for face generation: A survey. ACM Comput. Surv., 55(5):94:1–94:37, 2023. [72] Fatma Ben Aissa, Mahmoud Mejdoub, and Mourad Zaied. A survey on generative adversarial networks and their variants methods.In Wolfgang Osten and Dmitry P. Nikolaev, editors,Twelfth International Conference on Machine Vision, ICMV 2019, Amsterdam, The Nether- lands, 16-18 November 2019, volume 11433 ofSPIE Proceedings, page 114333N. SPIE, 2019. [73] Gustavo Henrique de Rosa and Jo ̃ ao P. Papa.A survey on text generation using generative adversarial networks.Pattern Recognit., 119:108098, 2021. [74] Rick Sauber-Cole and Taghi M. Khoshgoftaar. The use of generative adversarial networks to alleviate class imbalance in tabular data: a survey.J. Big Data, 9(1):98, 2022. [75] Vignesh Sampath, I ̃ naki Maurtua, Juan Jos ́ e Aguilar Mart ́ ın, and Aitor Gutierrez. A survey on generative adversarial networks for imbalance problems in computer vision tasks.J. Big Data, 8(1):27, 2021. [76] Qiuhua Yi, Xiangfan Chen, Chenwei Zhang, Zehai Zhou, Linan Zhu, and Xiangjie Kong. Diffusion models in text generation: a survey. PeerJ Comput. Sci., 10:e1905, 2024. [77] Yifan Li, Kun Zhou, Wayne Xin Zhao, and Ji-Rong Wen. Diffusion models for non-autoregressive text generation: A survey. InProceed- ings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI 2023, 19th-25th August 2023, Macao, SAR, China, pages 6692–6701. ijcai.org, 2023. [78] Hanqun Cao, Cheng Tan, Zhangyang Gao, Yilun Xu, Guangyong Chen, Pheng-Ann Heng, and Stan Z Li. A survey on generative diffusion models.IEEE Transactions on Knowledge and Data Engineering, 2024. [79] Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli.Deep unsupervised learning using nonequilibrium thermodynamics. In Francis R. Bach and David M. Blei, editors, Proceedings of the 32nd International Conference on Machine Learn- ing, ICML 2015, Lille, France, 6-11 July 2015, volume 37 ofJMLR Workshop and Conference Proceedings, pages 2256–2265. JMLR.org, 2015. [80] Jonathan Ho, Ajay Jain, and Pieter Abbeel.Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. [81] Prafulla Dhariwal and Alexander Nichol.Diffusion models beat gans on image synthesis.Advances in neural information processing systems, 34:8780–8794, 2021. [82] Stephen Villanueva, Abu Saleh Md Tayeen, Qixu Gong, Satyajayant Misra, Aden Dogar, Huiping Cao, Jiefei Liu, Pratyay Kumar, and Jayashree Harikumar. Netif: Network traffic to image features for robust intrusion detection.InMILCOM 2025-2025 IEEE Military Communications Conference (MILCOM), pages 1–6. IEEE, 2025. [83] Li Yang and Abdallah Shami. A transfer learning and optimized cnn based intrusion detection system for internet of vehicles. InICC 2022- IEEE International Conference on Communications, pages 2774–2779. IEEE, 2022. [84] Hojjat Navidan, Parisa Fard Moshiri, Mohammad Nabati, Reza Shah- bazian, Seyed Ali Ghorashi, Vahid Shah-Mansouri, and David Win- dridge.Generative adversarial networks (gans) in networking: A comprehensive survey & evaluation.Comput. Networks, 194:108149, 2021. [85] Aeryn Dunmore, Julian Jang-Jaccard, Fariza Sabrina, and Jin Kwak. A comprehensive survey of generative adversarial networks (gans) in cybersecurity intrusion detection.IEEE Access, 11:76071–76094, 2023. [86] Chenhan Zhang, Shui Yu, Zhiyi Tian, and James J. Q. Yu. Generative adversarial networks: A survey on attack and defense perspective.ACM Comput. Surv., 56(4):91:1–91:35, 2024. [87] Indira Kalyan Dutta, Bhaskar Ghosh, Albert H. Carlson, Michael W. Totaro, and Magdy A. Bayoumi.Generative adversarial networks in security: A survey. In11th IEEE Annual Ubiquitous Computing, Electronics & Mobile Communication Conference, UEMCON 2020, New York City, NY, USA, October 28-31, 2020, pages 399–405. IEEE, 2020. [88] Weili Nie et al. Diffusion models for adversarial purification. In International Conference on Machine Learning (ICML), 2022. [89] Xinquan Chen et al. Advdiffuser: Natural adversarial example synthesis with diffusion models. InProceedings of the IEEE/CVF ICCV, pages 4562–4572, October 2023. [90] Pratyay Kumar, Abu Saleh Md Tayeen, Satyajayant Misra, Huiping Cao, Jiefei Liu, Qixu Gong, and Jayashree Harikumar. Netdiffuser: Deceiving dnn-based network attack detection systems with diffusion- generated adversarial traffic.arXiv preprint arXiv:2603.08901, 2026. [91] Ashish Hooda, Neal Mangaokar, Ryan Feng, Kassem Fawaz, Somesh Jha, and Atul Prakash. D4: detection of adversarial diffusion deepfakes using disjoint ensembles. InIEEE/CVF Winter Conference on Appli- 19 cations of Computer Vision, WACV 2024, Waikoloa, HI, USA, January 3-8, 2024, pages 3800–3810. IEEE, 2024. [92] Nicholas Carlini, Florian Tram ` er, Krishnamurthy (Dj) Dvijotham, Leslie Rice, Mingjie Sun, and J. Zico Kolter. (certified!!) adversarial robustness for free!InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net, 2023. [93] Minjong Lee and Dongwoo Kim. Robust evaluation of diffusion-based adversarial purification. InIEEE/CVF International Conference on Computer Vision, ICCV 2023, Paris, France, October 1-6, 2023, pages 134–144. IEEE, 2023. [94] Quanlin Wu, Hang Ye, and Yuntian Gu. Guided diffusion model for adversarial purification from random noise.CoRR, abs/2206.10875, 2022. [95] Yiming Liu, Kezhao Liu, Yao Xiao, Ziyi Dong, Xiaogang Xu, Pengxu Wei, and Liang Lin.Towards better adversarial purification via adversarial denoising diffusion training.CoRR, abs/2404.14309, 2024. [96] Erjin Bao, Ching-Chun Chang, Huy H. Nguyen, and Isao Echizen. From deconstruction to reconstruction: A plug-in module for diffusion- based purification of adversarial examples. In Bin Ma, Jian Li, and Qi Li, editors,Digital Forensics and Watermarking - 22nd International Workshop, IWDW 2023, Jinan, China, November 25-26, 2023, Revised Selected Papers, volume 14511 ofLecture Notes in Computer Science, pages 48–62. Springer, 2023. [97] Jinyi Wang, Zhaoyang Lyu, Dahua Lin, Bo Dai, and Hongfei Fu.Guided diffusion model for adversarial purification.CoRR, abs/2205.14969, 2022. [98] Jiachen Sun, Weili Nie, Zhiding Yu, Z. Morley Mao, and Chaowei Xiao. Pointdp: Diffusion-driven purification against adversarial attacks on 3d point cloud recognition.CoRR, abs/2208.09801, 2022. [99] Mert Gulsen, Batuhan Cengiz, Yusuf Huseyin Sahin, and Gozde Unal. PCLD: point cloud layerwise diffusion for adversarial purification. CoRR, abs/2403.06698, 2024. [100] Xintian Ren, Jun Wu, Hansong Xu, and Xiuzhen Chen. Diffusion model based secure semantic communications with adversarial purifi- cation. In10th IEEE Conference on Big Data Security on Cloud, BigDataSecurity 2024, New York City, NY, USA, May 10-12, 2024, pages 130–134. IEEE, 2024. [101] Guang Lin, Zerui Tao, Jianhai Zhang, Toshihisa Tanaka, and Qibin Zhao. Robust diffusion models for adversarial purification.CoRR, abs/2403.16067, 2024. [102] Yibo Bai and Xiao-Lei Zhang. Diffusion-based adversarial purification for speaker verification.CoRR, abs/2310.14270, 2023. [103] Yang Song, Taesup Kim, Sebastian Nowozin, Stefano Ermon, and Nate Kushman. Pixeldefend: Leveraging generative models to under- stand and defend against adversarial examples. In6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings. OpenReview.net, 2018. [104] Mintong Kang, Dawn Song, and Bo Li. Diffattack: Evasion attacks against diffusion-based adversarial purification. In Alice Oh, Tristan Naumann, Amir Globerson, Kate Saenko, Moritz Hardt, and Sergey Levine, editors,Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023, 2023. [105] Weili Nie, Brandon Guo, Yujia Huang, Chaowei Xiao, Arash Vah- dat, and Animashree Anandkumar. Diffusion models for adversarial purification.In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csaba Szepesv ́ ari, Gang Niu, and Sivan Sabato, editors,International Conference on Machine Learning, ICML 2022, 17-23 July 2022, Baltimore, Maryland, USA, volume 162 ofProceedings of Machine Learning Research, pages 16805–16827. PMLR, 2022. [106] Jongmin Yu, Hyeontaek Oh, and Jinhong Yang.Adversarial de- noising diffusion model for unsupervised anomaly detection.CoRR, abs/2312.04382, 2023. [107] Ludwig Schmidt, Shibani Santurkar, Dimitris Tsipras, Kunal Talwar, and Aleksander Madry. Adversarially robust generalization requires more data. In Samy Bengio, Hanna M. Wallach, Hugo Larochelle, Kristen Grauman, Nicol ` o Cesa-Bianchi, and Roman Garnett, editors, Advances in Neural Information Processing Systems 31: Annual Con- ference on Neural Information Processing Systems 2018, NeurIPS 2018, December 3-8, 2018, Montr ́ eal, Canada, pages 5019–5031, 2018. [108] David Stutz, Matthias Hein, and Bernt Schiele. Disentangling adver- sarial robustness and generalization. InIEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019, Long Beach, CA, USA, June 16-20, 2019, pages 6976–6987. Computer Vision Foundation / IEEE, 2019. [109] Yinpeng Dong, Zhijie Deng, Tianyu Pang, Jun Zhu, and Hang Su. Adversarial distributional training for robust deep learning. In Hugo Larochelle, Marc’Aurelio Ranzato, Raia Hadsell, Maria-Florina Balcan, and Hsuan-Tien Lin, editors,Advances in Neural Information Process- ing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual, 2020. [110] Huaxia Wang and Chun-Nam Yu. A direct approach to robust deep learning using adversarial networks. In7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net, 2019. [111] Haoming Jiang, Zhehui Chen, Yuyang Shi, Bo Dai, and Tuo Zhao. Learning to defend by learning to attack.In Arindam Banerjee and Kenji Fukumizu, editors,The 24th International Conference on Artificial Intelligence and Statistics, AISTATS 2021, April 13-15, 2021, Virtual Event, volume 130 ofProceedings of Machine Learning Re- search, pages 577–585. PMLR, 2021. [112] Zekai Wang, Tianyu Pang, Chao Du, Min Lin, Weiwei Liu, and Shuicheng Yan. Better diffusion models further improve adversarial training.In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors, International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, volume 202 ofProceedings of Machine Learning Research, pages 36246–36263. PMLR, 2023. [113] Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adversarial examples. In Yoshua Bengio and Yann LeCun, editors,3rd International Conference on Learning Represen- tations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. [114] Bin Tang, Yan Lu, Qi Li, Yueying Bai, Jie Yu, and Xu Yu. A diffusion model based on network intrusion detection method for industrial cyber-physical systems.Sensors, 23(3):1141, 2023. [115] Yingshu Wang, Jian Ding, Xiaomeng He, Qian Wei, Shu Yuan, and Jun Zhang. Intrusion detection method based on denoising diffusion prob- abilistic models for uav networks.Mobile Networks and Applications, pages 1–10, 2023. [116] Soohyung Lee, Beomseong Kim, and Heesung Lee. Data augmentation using generative models for track intrusion detection.Science Progress, 106(4):00368504231212769, 2023. [117] Emiel Hoogeboom, Didrik Nielsen, Priyank Jaini, Patrick Forr ́ e, and Max Welling.Argmax flows and multinomial diffusion: Learning categorical distributions. In Marc’Aurelio Ranzato, Alina Beygelzimer, Yann N. Dauphin, Percy Liang, and Jennifer Wortman Vaughan, edi- tors,Advances in Neural Information Processing Systems 34: Annual Conference on Neural Information Processing Systems 2021, NeurIPS 2021, December 6-14, 2021, virtual, pages 12454–12465, 2021. [118] Mao Ye, Lemeng Wu, and Qiang Liu. First hitting diffusion models for generating manifold, graph and categorical data. In Sanmi Koyejo, S. Mohamed, A. Agarwal, Danielle Belgrave, K. Cho, and A. Oh, ed- itors,Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022, 2022. [119] Ting Chen, Ruixiang Zhang, and Geoffrey E. Hinton. Analog bits: Generating discrete data using diffusion models with self-conditioning. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenReview.net, 2023. [120] Akim Kotelnikov, Dmitry Baranchuk, Ivan Rubachev, and Artem Babenko. Tabddpm: Modelling tabular data with diffusion models. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara En- gelhardt, Sivan Sabato, and Jonathan Scarlett, editors,International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Hon- olulu, Hawaii, USA, volume 202 ofProceedings of Machine Learning Research, pages 17564–17579. PMLR, 2023. [121] Jiefei Liu, Qixu Gong, Wenbin Jiang, Pratyay Kumar, Abu Saleh Md Tayeen, Huiping Cao, Satyajayant Misra, and Jayashree Harikumar. Is synthetic flow data from generative models ready for network intrusion detection systems? InMILCOM 2025-2025 IEEE Military Communications Conference (MILCOM), pages 1–8. IEEE, 2025. [122] Oscar G. Lira, Alberto Marroquin, and Marco Antonio To. Harnessing the advanced capabilities of LLM for adaptive intrusion detection systems. In Leonard Barolli, editor,Advanced Information Networking and Applications - Proceedings of the 38th International Conference on Advanced Information Networking and Applications (AINA-2024), Kitakyushu, Japan, 17-19 April, 2024, Volume 6, volume 204 ofLecture Notes on Data Engineering and Communications Technologies, pages 453–464. Springer, 2024. 20 [123] Liam Daly Manocchio, Siamak Layeghy, Wai Weng Lo, Gayan K. Kulatilleke, Mohanad Sarhan, and Marius Portmann. Flowtransformer: A transformer framework for flow-based network intrusion detection systems.Expert Syst. Appl., 241:122564, 2024. [124] Jinhee Kim, Taesung Kim, and Jaegul Choo. Exploring prompting methods for mitigating class imbalance through synthetic data genera- tion with large language models, 2024. [125] Nabeel Seedat, Nicolas Huynh, Boris van Breugel, and Mihaela van der Schaar. Curated llm: Synergy of llms and data curation for tabular aug- mentation in ultra low-data regimes.arXiv preprint arXiv:2312.12112, 2023. [126] Yael Einy, Tova Milo, and Slava Novgorodov.Cost-effective llm utilization for machine learning tasks over tabular data. InProceedings of the Conference on Governance, Understanding and Integration of Data for Effective and Responsible AI, pages 45–49, 2024. [127] Shengzhe Xu, Cho-Ting Lee, Mandar Sharma, Raquib Bin Yousuf, Nikhil Muralidhar, and Naren Ramakrishnan. Are llms naturally good at synthetic tabular data generation?arXiv preprint arXiv:2406.14541, 2024. [128] Gaith Rjoub, Jamal Bentahar, Omar Abdel Wahab, Rabeb Mizouni, Alyssa Song, Robin Cohen, Hadi Otrok, and Azzam Mourad.A survey on explainable artificial intelligence for cybersecurity.IEEE Transactions on Network and Service Management, 20(4):5115–5140, 2023. [129] Victor J ̈ uttner, Martin Grimmer, and Erik Buchmann.Chatids: Explainable cybersecurity using generative ai.arXiv preprint arXiv:2306.14504, 2023. [130] Tarek Ali and Panos Kostakos. Huntgpt: Integrating machine learning- based anomaly detection and explainable ai with large language models (llms).arXiv preprint arXiv:2309.16021, 2023. [131] Dattaraj Rao and Shraddha Mane.Zero-shot learning approach to adaptive cybersecurity using explainable ai.arXiv preprint arXiv:2106.14647, 2021. [132] Chen Zhong, Qinwei Ni, and Ping Chen. Predicting analysts’ needs for explainable artificial intelligence (XAI) in cybersecurity analysis. In Paul A. Pavlou, Vishal Midha, Animesh Animesh, Traci A. Carte, Alexandre R. Graeml, and Alanah Mitchell, editors,29th Americas Conference on Information Systems, AMCIS 2023, Panama City, Panama, August 10-12, 2023. Association for Information Systems, 2023. [133] Toan V Tran and Li Xiong. Differentially private tabular data synthesis using large language models.arXiv preprint arXiv:2406.01457, 2024. [134] Durk P Kingma, Tim Salimans, Rafal Jozefowicz, Xi Chen, Ilya Sutskever, and Max Welling.Improved variational inference with inverse autoregressive flow.Advances in neural information processing systems, 29, 2016. [135] Martin Arjovsky and L ́ eon Bottou.Towards principled meth- ods for training generative adversarial networks.arXiv preprint arXiv:1701.04862, 2017. [136] Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative adversarial networks. In5th International Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017, Conference Track Proceedings. OpenReview.net, 2017. [137] Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. InInternational conference on machine learning, pages 8162–8171. PMLR, 2021. [138] Solomon Kullback and Richard A Leibler.On information and sufficiency.The annals of mathematical statistics, 22(1):79–86, 1951. [139] Jianhua Lin. Divergence measures based on the shannon entropy.IEEE Transactions on Information theory, 37(1):145–151, 1991. [140] C ́ edric Villani et al.Optimal transport: old and new, volume 338. Springer, 2009. [141] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter.Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017. [142] Arthur Gretton, Karsten M Borgwardt, Malte J Rasch, Bernhard Sch ̈ olkopf, and Alexander Smola. A kernel two-sample test.The Journal of Machine Learning Research, 13(1):723–773, 2012. [143] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4401–4410, 2019. [144] G ́ abor J Sz ́ ekely and Maria L Rizzo.Energy statistics: A class of statistics based on distances.Journal of statistical planning and inference, 143(8):1249–1272, 2013. [145] Mehdi SM Sajjadi, Olivier Bachem, Mario Lucic, Olivier Bousquet, and Sylvain Gelly. Assessing generative models via precision and recall.Advances in neural information processing systems, 31, 2018. [146] Maximilian Schmidt, Andrea Bartezzaghi, and Ngoc Thang Vu. Prompting-based synthetic data generation for few-shot question an- swering.arXiv preprint arXiv:2405.09335, 2024. [147] Giulia DeSalvo, Jean-Fracois Kagy, Lazaros Karydas, Afshin Ros- tamizadeh, and Sanjiv Kumar. No more hard prompts: Softsrv prompt- ing for synthetic data generation.arXiv preprint arXiv:2410.16534, 2024. [148] Yufei Wang, Can Xu, Qingfeng Sun, Huang Hu, Chongyang Tao, Xiubo Geng, and Daxin Jiang. Promda: Prompt-based data augmentation for low-resource NLU tasks. In Smaranda Muresan, Preslav Nakov, and Aline Villavicencio, editors,Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2022, Dublin, Ireland, May 22-27, 2022, pages 4242– 4255. Association for Computational Linguistics, 2022. [149] Fiona Victoria Stanley Jothiraj and Afra Mashhadi.Phoenix: A federated generative diffusion model. In Tat-Seng Chua, Chong-Wah Ngo, Roy Ka-Wei Lee, Ravi Kumar, and Hady W. Lauw, editors, Companion Proceedings of the ACM on Web Conference 2024, W 2024, Singapore, Singapore, May 13-17, 2024, pages 1568–1577. ACM, 2024. [150] Nguyen Chi Vy, Nguyen Huu Quyen, Phan The Duy, and Van-Hau Pham.Federated learning-based intrusion detection in the context of iiot networks: Poisoning attack and defense. In Min Yang, Chao Chen, and Yang Liu, editors,Network and System Security - 15th International Conference, NSS 2021, Tianjin, China, October 23, 2021, Proceedings, volume 13041 ofLecture Notes in Computer Science, pages 131–147. Springer, 2021. [151] Aliya Tabassum, Aiman Erbad, Wadha Lebda, Amr Mohamed, and Mohsen Guizani. FEDGAN-IDS: privacy-preserving IDS using GAN and federated learning.Comput. Commun., 192:299–310, 2022. [152] Zilong Lin, Yong Shi, and Zhi Xue. IDSGAN: generative adversarial networks for attack generation against intrusion detection. In Jo ̃ ao Gama, Tianrui Li, Yang Yu, Enhong Chen, Yu Zheng, and Fei Teng, editors,Advances in Knowledge Discovery and Data Mining - 26th Pacific-Asia Conference, PAKDD 2022, Chengdu, China, May 16- 19, 2022, Proceedings, Part I, volume 13282 ofLecture Notes in Computer Science, pages 79–91. Springer, 2022. [153] Yuemeng Zhang, Yong Zhang, Zhao Zhang, Haonan Bai, Tianyi Zhong, and Mei Song. Evaluation of data poisoning attacks on federated learning-based network intrusion detection system. In24th IEEE Int Conf on High Performance Computing & Communications; 8th Int Conf on Data Science & Systems; 20th Int Conf on Smart City; 8th Int Conf on Dependability in Sensor, Cloud & Big Data Systems & Application, HPCC/DSS/SmartCity/DependSys 2022, Hainan, China, December 18-20, 2022, pages 2235–2242. IEEE, 2022. [154] Malathy N, Shree Harish Kumar G, Sriram R, and Jebocen Im- manuel Raj NR. Federated transfer learning for intrusion detection system in industrial iot 4.0.Multim. Tools Appl., 83(19):57913–57941, 2024. [155] Pramod Munaweera, Suranga Prasad, Tharaka Hewa, Yushan Siri- wardhana, and Mika Ylianttila.Federated learning-powered ddos attack detection for securing cyber physical systems in 5g and beyond networks. InProceedings of the 14th International Conference on the Internet of Things, pages 273–278, 2024. [156] Jiefei Liu, Huiping Cao, Abu Saleh Md Tayeen, Satyajayant Misra, Pratyay Kumar, and Jayashree Harikumar. Multi-model-based federated learning to overcome local class imbalance issues. In2023 Interna- tional Conference on Machine Learning and Applications (ICMLA), pages 265–270. IEEE, 2023. [157] Chao Zha, Zhiyu Wang, Yifei Fan, Xingming Zhang, Bing Bai, Yinjie Zhang, Sainan Shi, and Ruyun Zhang.Skt-ids: Unknown attack detection method based on sigmoid kernel transformation and encoder– decoder architecture.Computers & Security, 146:104056, 2024. [158] Ruming Tang, Zheng Yang, Zeyan Li, Weibin Meng, Haixin Wang, Qi Li, Yongqian Sun, Dan Pei, Tao Wei, Yanfei Xu, et al. Zerowall: Detecting zero-day web attacks through encoder-decoder recurrent neural networks.InIEEE INFOCOM 2020-IEEE Conference on Computer Communications, pages 2479–2488. IEEE, 2020. [159] Yisroel Mirsky, Tomer Doitshman, Yuval Elovici, and Asaf Shabtai. Kitsune: an ensemble of autoencoders for online network intrusion detection.arXiv preprint arXiv:1802.09089, 2018. [160] Abu Saleh Md Tayeen, Satyajayant Misra, Huiping Cao, and Jayashree Harikumar. Cafnet: Compressed autoencoder-based federated network 21 for anomaly detection. InMILCOM 2023-2023 IEEE Military Com- munications Conference (MILCOM), pages 325–330. IEEE, 2023. [161] Muyang Li, Ji Lin, Yaoyao Ding, Zhijian Liu, Jun-Yan Zhu, and Song Han. Gan compression: Efficient architectures for interactive conditional gans.InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5284–5294, 2020. [162] Xiaowei Hu, Zhe Gan, Jianfeng Wang, Zhengyuan Yang, Zicheng Liu, Yumao Lu, and Lijuan Wang. Scaling up vision-language pre-training for image captioning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 17980–17989, 2022. [163] Haomiao Yang, Mengyu Ge, Kunlan Xiang, Xuejun Bai, and Hongwei Li. Fedvae: Communication-efficient federated learning with non-iid private data.IEEE Systems Journal, 17(3):4798–4808, 2023. [164] Mahdi Morafah, Matthias Reisser, Bill Lin, and Christos Louizos. Stable diffusion-based data augmentation for federated learning with non-iid data.arXiv preprint arXiv:2405.07925, 2024. [165] Jianqi Liu, Zhiwei Zhao, Xiangyang Luo, Pan Li, Geyong Min, and Huiyong Li. Slaugfl: Efficient edge federated learning with selective gan-based data augmentation.IEEE Transactions on Mobile Comput- ing, 2024. [166] Jayneel Vora, Nader Bouacida, Aditya Krishnan, and Prasant Mo- hapatra. Feddm: Enhancing communication efficiency and handling data heterogeneity in federated diffusion models.arXiv preprint arXiv:2407.14730, 2024. [167] Seyoung Ahn, Soohyeong Kim, Yongseok Kwon, Joohan Park, Jiseung Youn, and Sunghyun Cho. Communication-efficient diffusion strategy for performance improvement of federated learning with non-iid data. arXiv preprint arXiv:2207.07493, 2022.