ASSET

Abstract:  Machine learning applications are increasingly reliant on black-box pretrained models. To ensure safe use of these models, techniques such as unlearning, guardrails, and watermarking have been proposed to curb model behavior and audit usage. Unfortunately, while these post-hoc approaches give positive safety ‘vibes’ when evaluated in isolation, our work shows that existing techniques are quite brittle when deployed […]

Abstract: Large Language Models (LLMs) are vulnerable to adversarial attacks, which bypass common safeguards put in place to prevent these models from generating harmful output. Notably, these attacks can be transferrable to other models---even proprietary ones—potentially compromising a wide range of AI systems with a single exploit. This surprising fragility underscores a critical weakness in […]

Abstract: Robust simulation and precise modeling of physical dynamics are essential for advancing perception, planning, and control in the development of generalist physical agents. In this talk, I will present my research on building generative models that combine physical realism with scalability in high-dimensional environments. The presentation delves into both the theoretical foundations and practical […]

Abstract: American democracy has been undermined by an “infodemic” of fake news, coupled with the widespread segregation of consumers into ideologically homogenous echo chambers by inscrutable algorithms deployed by rapacious social media platforms—or so we are told. In this talk, I will critically examine claims of this sort—made frequently by politicians, journalists, and public intellectuals—summarizing […]

Abstract: Neurosymbolic Program Synthesis (NSP) integrates neural networks and symbolic reasoning to tackle complex tasks requiring both perception and logical reasoning. This talk provides an overview of the NSP framework and its applications in domains such as image editing, data extraction, and robot learning from demonstrations. We will delve into the key ideas behind NSP […]