Virtual reality (VR) platforms enable a wide range of applications, however, pose unique privacy risks. In particular, VR devices are equipped with a rich set of sensors that collect personal and sensitive information (e.g., body motion, eye gaze, hand joints, and facial expression). The data from these newly available sensors can be used to uniquely identify a user, even in the absence of explicit identifiers. In this paper, we seek to understand the extent to which a user can be identified based solely on VR sensor data, within and across real-world apps from diverse genres. We consider adversaries with capabilities that range from observing APIs available within a single app (app adversary) to observing all or selected sensor measurements across multiple apps on the VR device (device adversary). To that end, we introduce BehaVR, a framework for collecting and analyzing data from all sensor groups collected by multiple apps running on a VR device. We use BehaVR to collect data from real users that interact with 20 popular real-world apps. We use that data to build machine learning models for user identification within and across apps, with features extracted from available sensor data. We show that these models can identify users with an accuracy of up to 100%, and we reveal the most important features and sensor groups, depending on the functionality of the app and the adversary. To the best of our knowledge, BehaVR is the first to analyze user identification in VR comprehensively, i.e., considering all sensor measurements available on consumer VR devices, collected by multiple real-world, as opposed to custom-made, apps.

In membership inference attacks (MIAs), an adversary observes the predictions of a model to determine whether a sample is part of the model’s training data. Existing MIA defenses conceal the presence of a target sample through strong regularization, knowledge distillation, confidence masking, or differential privacy. We propose MIAShield, a new MIA defense based on preemptive exclusion of member samples instead of masking the presence of a member. MIAShield departs from prior defenses in that it weakens the strong membership signal that stems from the presence of a target sample by preemptively excluding it at prediction time without compromising model utility. To that end, we design and evaluate a suite of preemptive exclusion oracles leveraging model confidence, exact/approximate sample signature, and learning-based exclusion of member data points. To be practical, MIAShield splits a training data into disjoint subsets and trains each subset to build an ensemble of models. The disjointedness of subsets ensures that a target sample belongs to only one subset, which isolates the sample to facilitate the preemptive exclusion goal. We evaluate MIAShield on three benchmark image classification datasets. We show that MIAShield effectively mitigates membership inference (near random guess) for a wide range of MIAs; achieves far better privacy/utility trade-off compared with state-of-the-art defenses; and remains resilient in the face of adaptive attacks.

Differential Privacy (DP) has emerged as a rigorous formalism to quantify privacy protection provided by an algorithm that operates on privacy sensitive data. In machine learning (ML), DP has been employed to limit inference/disclosure of training examples. Prior work leveraged DP across the ML pipeline, albeit in isolation, often focusing on mechanisms such as gradient perturbation.