Risk-sensitive reinforcement learning (RL) with an entropic risk measure typically requires knowledge of the transition kernel or performs unstable updates w.r.t. exponential Bellman equations. As a consequence, algorithms that optimize this objective have been restricted to tabular or low-dimensional continuous environments. In this work we leverage the connection between the entropic risk measure and the RL-as-inference framework to develop a risk-sensitive variational actor-critic algorithm (rsVAC). Our work extends the variational framework to incorporate stochastic rewards and proposes a variational model-based actor-critic approach that modulates policy risk via a risk parameter. We consider, both, the risk-seeking and risk-averse regimes and present rsVAC learning variants for each setting. Our experiments demonstrate that this approach produces risk-sensitive policies and yields improvements in both tabular and risk-aware variants of complex continuous control tasks in MuJoCo.

Deep learning (DL)-based malware detectors have shown promise in swiftly detecting unseen malware without expensive dynamic malware behavior analysis. These detectors have been shown to be susceptible to adversarial malware variants generated from meticulously modifying known malware to mislead detectors into recognizing them as benign. Being able to automatically generate optimized functional adversarial malware variants by defenders is crucial to effective cyber defense and staying ahead of the adversary. Current adversarial malware example generation methods often assume threat models with any of the following four restrictions: (1) requiring access to insider knowledge about malware detectors, (2) an unlimited size of adversarial modifications, (3) an unlimited number of queries to malware detector, and (4) relying on dynamic analysis of malware behavior in a sandbox. Drawing on Actor-Critic Reinforcement Learning (RL), we propose a novel closed-box binary manipulation method for adversarial malware optimization, named Actor-Critic with Contextualized Action Representations (AC-CAR), to generate malware variants without these restrictions. AC-CAR leverages two novel components, a contextualized policy and a neural language model-based RL-augmented top-$k$sampling method. Unlike current methods, AC-CAR can utilize tens of thousands of actions to augment malware executables for evading DL-based malware detectors. AC-CAR yields an approximately 2-fold performance increase over the current methods on average, while decreasing the payload size to 20 times smaller than leading methods. We show that using the malware variants generated by AC-CAR in an adversarial re-training procedure improves malware detector’ robustness against adversarial variants by 29.65% on average.

Federated learning (FL) enables collaborative model training while preserving user data privacy by keeping data local. Despite these advantages, FL remains vulnerable to privacy attacks on user updates and model parameters during training and deployment. Secure aggregation protocols have been proposed to protect user updates by encrypting them, but these methods often incur high computational costs and are not resistant to quantum computers. Additionally, differential privacy (DP) has been used to mitigate privacy leakages, but existing methods focus on secure aggregation or DP, neglecting their potential synergies. To address these gaps, we introduce$\texttt {Beskar}$, a novel framework that provides post-quantum secure aggregation, optimizes computational overhead for FL settings, and defines a comprehensive threat model that accounts for a wide spectrum of adversaries. We also integrate DP into different stages of FL training to enhance privacy protection in diverse scenarios. Our framework provides a detailed analysis of the trade-offs between security, performance, and model accuracy, representing the first thorough examination of secure aggregation protocols combined with various DP approaches for post-quantum secure FL.$\texttt {Beskar}$aims to address the pressing privacy and security issues FL while ensuring quantum-safety and robust performance.

Secure aggregation protocols ensure the privacy of users’ data in federated learning by preventing the disclosure of local gradients. Many existing protocols impose significant communication and computational burdens on participants and may not efficiently handle the large update vectors typical of machine learning models. Correspondingly, we present e-SeaFL, an efficient verifiable secure aggregation protocol taking only one communication round during the aggregation phase. e-SeaFL allows the aggregation server to generate proof of honest aggregation to participants via authenticated homomorphic vector commitments. Our core idea is the use of assisting nodes to help the aggregation server, under similar trust assumptions existing works place upon the participating users. Our experiments show that the user enjoys an order of magnitude efficiency improvement over the state-of-the-art (IEEE S&P 2023) for large gradient vectors with thousands of parameters. Our open-source implementation is available at https://github.com/vt-asaplab/e-SeaFL.

Differentially private stochastic gradient descent (DP-SGD) has been instrumental in privately training deep learning models by providing a framework to control and track the privacy loss incurred during training. At the core of this computation lies a subsampling method that uses a privacy amplification lemma to enhance the privacy guarantees provided by the additive noise. Fixed size subsampling is appealing for its constant memory usage, unlike the variable sized minibatches in Poisson subsampling. It is also of interest in addressing class imbalance and federated learning. Current computable guarantees for fixed-size subsampling are not tight and do not consider both add/remove and replace-one adjacency relationships. We present a new and holistic Rényi differential privacy (RDP) accountant for DP-SGD with fixed-size subsampling without replacement (FSwoR) and with replacement (FSwR). For FSwoR we consider both add/remove and replace-one adjacency, where we improve on the best current computable bound by a factor of $4$. We also show for the first time that the widely-used Poisson subsampling and FSwoR with replace-one adjacency have the same privacy to leading order in the sampling probability. Our work suggests that FSwoR is often preferable to Poisson subsampling due to constant memory usage. Our FSwR accountant includes explicit non-asymptotic upper and lower bounds and, to the authors' knowledge, is the first such RDP analysis of fixed-size subsampling with replacement for DP-SGD. We analytically and empirically compare fixed size and Poisson subsampling, and show that DP-SGD gradients in a fixed-size subsampling regime exhibit lower variance in practice in addition to memory usage benefits.

Adversarial Malware Generation (AMG), the generation of adversarial malware variants to strengthen Deep Learning (DL)-based malware detectors has emerged as a crucial tool in the development of proactive cyberdefense. However, the majority of extant works offer subtle perturbations or additions to executable files and do not explore full-file obfuscation. In this study, we show that an open-source encryption tool coupled with a Reinforcement Learning (RL) framework can successfully obfuscate malware to evade state-of-the-art malware detection engines and outperform techniques that use advanced modification methods. Our results show that the proposed method improves the evasion rate from 27%-49% compared to widely-used state-of-the-art reinforcement learning-based methods.

Learning predictive models in new domains with scarce training data is a growing challenge in modern supervised learning scenarios. This incentivizes developing domain adaptation methods that leverage the knowledge in known domains (source) and adapt to new domains (target) with a different probability distribution. This becomes more challenging when the source and target domains are in heterogeneous feature spaces, known as heterogeneous domain adaptation (HDA). While most HDA methods utilize mathematical optimization to map source and target data to a common space, they suffer from low transferability. Neural representations have proven to be more transferable; however, they are mainly designed for homogeneous environments. Drawing on the theory of domain adaptation, we propose a novel framework, Heterogeneous Adversarial Neural Domain Adaptation (HANDA), to effectively maximize the transferability in heterogeneous environments. HANDA conducts feature and distribution alignment in a unified neural network architecture and achieves domain invariance through adversarial kernel learning. Three experiments were conducted to evaluate the performance against the state-of-the-art HDA methods on major image and text e-commerce benchmarks. HANDA shows statistically significant improvement in predictive performance. The practical utility of HANDA was shown in real-world dark web online markets. HANDA is an important step towards successful domain adaptation in e-commerce applications.

Deep Learning (DL)-based malware detectors are increasingly adopted for early detection of malicious behavior in cybersecurity. However, their sensitivity to adversarial malware variants has raised immense security concerns. Generating such adversarial variants by the defender is crucial to improving the resistance of DL-based malware detectors against them. This necessity has given rise to an emerging stream of machine learning research, Adversarial Malware example Generation (AMG), which aims to generate evasive adversarial malware variants that preserve the malicious functionality of a given malware. Within AMG research, black-box method has gained more attention than white-box methods. However, most black-box AMG methods require numerous interactions with the malware detectors to generate adversarial malware examples. Given that most malware detectors enforce a query limit, this could result in generating non-realistic adversarial examples that are likely to be detected in practice due to lack of stealth. In this study, we show that a novel DL-based causal language model enables single-shot evasion (i.e., with only one query to malware detector) by treating the content of the malware executable as a byte sequence and training a Generative Pre-Trained Transformer (GPT). Our proposed method, MalGPT, significantly outperformed the leading benchmark methods on a real-world malware dataset obtained from VirusTotal, achieving over 24.51% evasion rate. MalGPT enables cybersecurity researchers to develop advanced defense capabilities by emulating large-scale realistic AMG.

Internet users have been exposing an increasing amount of Personally Identifiable Information (PII) on social media. Such exposed PII can be exploited by cybercriminals and cause severe losses to the users. Informing users of their PII exposure in social media is crucial to raise their privacy awareness and encourage them to take protective measures. To this end, advanced techniques are needed to extract users’ exposed PII in social media automatically, whereas most existing studies remain manual. While Information Extraction (IE) techniques can be used to extract the PII automatically, Deep Learning (DL)-based IE models alleviate the need for feature engineering and further improve the efficiency. However, DL-based IE models often require large-scale labeled data for training, but PII-labeled social media posts are difficult to obtain due to privacy concerns. Also, these models rely heavily on pre-trained word embeddings, while PII in social media often varies in forms and thus has no fixed representations in pre-trained word embeddings. In this study, we propose the Deep Transfer Learning for PII Extraction (DTL-PIIE) framework to address these two limitations. DTL-PIIE transfers knowledge learned from publicly available PII data to social media in order to address the problem of rare PII-labeled data. Moreover, our framework leverages Graph Convolutional Networks (GCNs) to incorporate syntactic patterns to guide PIIE without relying on pre-trained word embeddings. Evaluation against benchmark IE models indicates that our approach outperforms state-of-the-art DL-based IE models. An ablation analysis further confirms the efficacy of each component in our model. Our proposed framework can facilitate various applications, such as PII misuse prediction and privacy risk assessment, thereby protecting the privacy of internet users.

Personally identifiable information (PII) has become a major target of cyber-attacks, causing severe losses to data breach victims. To protect data breach victims, researchers focus on collecting exposed PII to assess privacy risk and identify at-risk individuals. However, existing studies mostly rely on exposed PII collected from either the dark web or the surface web. Due to the wide exposure of PII on both the dark web and surface web, collecting from only the dark web or the surface web could result in an underestimation of privacy risk. Despite its research and practical value, jointly collecting PII from both sources is a non-trivial task. In this paper, we summarize our effort to systematically identify, collect, and monitor a total of 1,212,004,819 exposed PII records across both the dark web and surface web. Our effort resulted in 5.8 million stolen SSNs, 845,000 stolen credit/debit cards, and 1.2 billion stolen account credentials. From the surface web, we identified and collected over 1.3 million PII records of the victims whose PII is exposed on the dark web. To the best of our knowledge, this is the largest academic collection of exposed PII, which, if properly anonymized, enables various privacy research inquiries, including assessing privacy risk and identifying at-risk populations.

Cyber threat intelligence (CTI) necessitates automated monitoring of dark web platforms (e.g., Dark Net Markets and carding shops) on a large scale. While there are existing methods for collecting data from the surface web, large-scale dark web data collection is commonly hindered by anti-crawling measures. Text-based CAPTCHA serves as the most prohibitive type of these measures. Text-based CAPTCHA requires the user to recognize a combination of hard-to-read characters. Dark web CAPTCHA patterns are intentionally designed to have additional background noise and variable character length to prevent automated CAPTCHA breaking. Existing CAPTCHA breaking methods cannot remedy these challenges and are therefore not applicable to the dark web. In this study, we propose a novel framework for breaking text-based CAPTCHA in the dark web. The proposed framework utilizes Generative Adversarial Network (GAN) to counteract dark web-specific background noise and leverages an enhanced character segmentation algorithm. Our proposed method was evaluated on both benchmark and dark web CAPTCHA testbeds. The proposed method significantly outperformed the state-of-the-art baseline methods on all datasets, achieving over 92.08% success rate on dark web testbeds. Our research enables the CTI community to develop advanced capabilities of large-scale dark web monitoring.

The frequency and costs of cyber-attacks are increasing each year. By the end of 2019, the total cost of data breaches is expected to reach $2.1 trillion through the ever-growing online presence of enterprises and their consumers. The tools to perform these attacks and the breached data can often be purchased within the Dark-net. Many of the threat actors within this realm use its various platforms to broker, discuss, and strategize these cyber-threat assets. To combat these attacks, researchers are developing Cyber-Threat Intelligence (CTI) tools to proactively monitor the ever-growing online hacker community. This paper will detail the creation and use of a CTI tool that leverages a social network to identify cyber-threats across major Dark-net data sources. Through this network, emerging threats can be quickly identified so proactive or reactive security measures can be implemented.

As the Internet based applications become more and more ubiquitous, drug retailing on Dark Net Marketplaces (DNMs) has raised public health and law enforcement concerns due to its highly accessible and anonymous nature. To combat illegal drug transaction among DNMs, authorities often require agents to impersonate DNM customers in order to identify key actors within the community. This process can be costly in time and resource. Research in DNMs have been conducted to provide better understanding of DNM characteristics and drug sellers' behavior. Built upon the existing work, researchers can further leverage predictive analytics techniques to take proactive measures and reduce the associated costs. To this end, we propose a systematic analytical approach to identify key opioid sellers in DNMs. Utilizing machine learning and text analysis, this research provides prediction of high-impact opioid products in two major DNMs. Through linking the high-impact products and their sellers, we then identify the key opioid sellers among the communities. This work intends to help law enforcement authorities to formulate strategies by providing specific targets within the DNMs and reduce the time and resources required for prosecuting and eliminating the criminals from the market.

Cyber-attacks cost the global economy over $450 billion annually. To combat this issue, researchers and practitioners put enormous efforts into developing Cyber Threat Intelligence, or the process of identifying emerging threats and key hackers. However, the reliance on internal network data to has resulted in inherently reactive intelligence. CTI experts have urged the importance of proactively studying the large, ever-evolving online hacker community. Despite their CTI value, collecting data from hacker community platforms is a non-trivial task. In this paper, we summarize our efforts in systematically identifying and automatically collecting a large-scale of hacker forums, carding shops, Internet-Relay-Chat, and Dark Net Marketplaces. We also present our efforts to provide this data to the larger CTI community via the AZSecure Hacker Assets Portal (www.azsecure-hap.com). With our methodology, we collected 102 platforms for a total of 43,981,647 records. To the best of our knowledge, this compilation of hacker community data is the largest such collection in academia.

Recent advances in proactive cyber threat intelligence rely on early detection of cyber threats in hacker communities. Dark Net Markets (DNMs) are growing platforms in hacker community that provide hackers with highly- specialized tools and products which may not be found in other platforms. While text classification techniques have been used for cyber threat detection in English DNMs, the task is hindered in non-English platforms due to the language barrier and lack of ground-truth data. Current approaches use monolingual models on machine translated data to overcome these challenges. However, the translation errors can deteriorate the classification results. The abundance of data in English DNMs can be leveraged in learning non-English threats without using machine translation. In this study, we show that a deep cross-lingual model that can jointly learn the common language representation from two languages, significantly outperforms a monolingual model learned on machine translated data for identifying cyber threats in non-English DNMs. Unlike most studies, our approach does not require any external data source such as bilingual word embeddings or bilingual lexicons. Our experiments on Russian DNMs show that this approach can achieve better performance than state-of-the-art methods for non-English cyber threat detection in malicious hacker community.