Social virtual reality is an emerging medium of communication. In this medium, a user’s avatar (virtual representation) is controlled by the tracked motion of the user’s headset and hand controllers. This tracked motion is a rich data stream that can leak characteristics of the user or can be effectively matched to previously-identified data to identify a user. To better understand the boundaries of motion data identifiability, we investigate how varying training data duration and train-test delay affects the accuracy at which a machine learning model can correctly classify user motion in a supervised learning task simulating re-identification. The dataset we use has a unique combination of a large number of participants, long duration per session, large number of sessions, and a long time span over which sessions were conducted. We find that training data duration and train-test delay affect identifiability; that minimal train-test delay leads to very high accuracy; and that train-test delay should be controlled in future experiments.

The use of virtual and augmented reality devices is increasing, but these sensor-rich devices pose risks to privacy. The ability to track a user’s motion and infer the identity or characteristics of the user poses a privacy risk that has received significant attention. Existing deep-network-based defenses against this risk, however, require significant amounts of training data and have not yet been shown to generalize beyond specific applications. In this work, we study the effect of signal degradation on identifiability, specifically through added noise, reduced framerate, reduced precision, and reduced dimensionality of the data. Our experiment shows that state-of-the-art identification attacks still achieve near-perfect accuracy for each of these degradations. This negative result demonstrates the difficulty of anonymizing this motion data and gives some justification to the existing data- and compute-intensive deep-network based methods.

The combination of strong visual backbones and Large Language Model (LLM) reasoning has led to Large Multimodal Models (LMMs) becoming the current standard for a wide range of vision and language (VL) tasks. However, recent research has shown that even the most advanced LMMs still struggle to capture aspects of compositional visual reasoning, such as attributes and relationships between objects. One solution is to utilize scene graphs (SGs)-a formalization of objects and their relations and attributes that has been extensively used as a bridge between the visual and textual domains. Yet, scene graph data requires scene graph annotations, which are expensive to collect and thus not easily scalable. Moreover, finetuning an LMM based on SG data can lead to catastrophic forgetting of the pretraining objective. To overcome this, inspired by chain-of-thought methods, we propose Compositional Chain-of-Thought (CCoT), a novel zero-shot Chain-of-Thought prompting method that utilizes SG representations in order to extract compositional knowledge from an LMM. Specifically, we first generate an SG using the LMM, and then use that SG in the prompt to produce a response. Through extensive experiments, we find that the proposed CCoT approach not only improves LMM performance on several vision and language (VL) compositional benchmarks but also improves the performance of several popular LMMs on general multimodal benchmarks, without the need for fine-tuning or annotated ground-truth SGs. Code: https://github.com/chancharikmitra/CCoT.

The recent success of interleaved Large Multimodal Models (LMMs) in few-shot learning suggests that in-context learning (ICL) with many examples can be promising for learning new tasks. However, this many-shot multimodal ICL setting has one crucial problem: it is fundamentally limited by the model's context length set at pretraining. The problem is especially prominent in the multimodal domain, which processes both text and images, requiring additional tokens. This motivates the need for a multimodal method to compress many shots into fewer tokens without finetuning. In this work, we enable LMMs to perform multimodal, many-shot in-context learning by leveraging Multimodal Task Vectors (MTV)---compact implicit representations of in-context examples compressed in the model's attention heads. Specifically, we first demonstrate the existence of such MTV in LMMs and then leverage these extracted MTV to enable many-shot in-context learning for various vision-and-language tasks. Our experiments suggest that MTV can scale in performance with the number of compressed shots and generalize to similar out-of-domain tasks without additional context length for inference. Code: https://github.com/Brandon3964/MultiModal-Task-Vector