Modern embedded systems are ever present within our daily lives. Such devices remain vulnerable to Differential Power Analysis via side-channel attacks, which provide a powerful and flexible framework for exfiltrating valuable information. This work applies these principles with respect to determining properties of the underlying finite state machine dictating code execution on a STM32F030 microcontroller. A K-nearest neighbors classification model is used to determine the order of state transitions within the finite state machine with a maximum accuracy of 81% over 256 different classes. A second experiment was performed to evaluate the effect of dimensionality on the classification accuracy and found an ability to classify with an accuracy over 95% in classes with 32 elements and perfectly classify between only two classes. Results validate the need for protections against power-based side-channel attacks when sensitive information is handled by FSM-controlled components.

Recent studies have shown Graph Neural Networks (GNNs) are extremely vulnerable to attribute inference attacks. To tackle this challenge, existing privacy-preserving GNNs research assumes that the sensitive attributes of all users are known beforehand. However, due to different privacy preferences, some users (i.e., private users) may prefer not to reveal sensitive information that others (i.e., non-private users) would not mind disclosing. For example, in social networks, male users are typically less sensitive to their age information than female users. The age disclosure of male users can lead to the age information of female users in the network exposed. This is partly because social media users are connected, the homophily property and message-passing mechanism of GNNs can exacerbate individual privacy leakage. In this work, we study a novel and practical problem of learning privacy-preserving GNNs with partially observed sensitive attributes.

Modern malware indicators utilized by the current top threat feeds are easily bypassed and generated through enigmatic methods, leading to a lack of detection capabilities for cyber defenders. Static hash-based algorithms such as MD5 or SHA generate indicators that are rendered obsolete by modifying a single byte of the source file. Conversely, fuzzy hash-based algorithms such as SSDEEP and TLSH are more robust to alterations of source information; however, these methods often utilize context boundaries that are hard to define or not based on meaningful information. In previous work, a custom binary analysis tool was created called @DisCo. In this study, four current ransomware campaigns were analyzed using TLSH fuzzy hashing and the @DisCo tool. While TLSH works on the binary level of the entire program, @DisCo works at an intermediate function level. The results from each analysis method were compared to provide validation between the two as well as introduce a narrative for using combinations of these types of methods for the creation of stronger indicators of compromise.

With the rising concerns over privacy and fairness in machine learning, privacy-preserving fair machine learning has received tremendous attention in recent years. However, most existing fair models still need to collect sensitive demographic data, which may be impossible given privacy regulations. To address the dilemma between model fairness and sensitive data collection, we propose DicPF, a distributed and privacy-preserving fair learning framework that operates without collecting sensitive demographic data. In particular, DicPF assumes multiple local agents and a modeler are distributed, and sensitive demographic data is separately held by multiple local agents. To assist fair learning at the modeler, each agent learns a fair local dictionary and send it to the modeler. The modeler learns a fair model based on an aggregated dictionary. Under DicPF framework, we propose a private z-Sparse Fair Learner. Extensive experiments on three real-world datasets demonstrate the efficiency of the proposed model. Compared with the state-of-the-art fair learners, the proposed z-Sparse Fair Learner achieves superior fairness performance by lowering prediction disparity. We also develop a privacy inference model to demonstrate the excellent privacy-preserving performance of DicPF. Finally, we theoretically analyze z-Sparse Fair Learner and prove upper bounds on its model fairness and accuracy.

Most current authentication mechanisms rely on static initial verification of the user; however, such authentication mechanisms do not verify user identities on already unlocked systems. Spy Hunter, a continuous authentication mechanism, constantly examines user's keystroke timing dynamics to assess the user's identity. Also, Spy Hunter preserves the privacy of the user by only utilizing the key press timings without storing information regarding which keys were pressed. Specifically, Spy Hunter implements two one-class classifiers, only utilizing genuine user samples for training the model, and mitigates adversarial inclusion of impostor data in the model. The data in this preliminary study consists of timing information from 20 users running Spy Hunter in the background while they used their systems. One-class classification was performed independently for all the users. Experimental results show that in less than 80 characters, Spy Hunter performed with as low as 2.05% False Acceptance Rate (FAR) and 2% False Rejection Rate (FRR).