ASSET

Abstract:  Extracting insights from imaging data used to be straightforward: every component of imaging systems was engineered by humans, the analysis and interpretation of the collected data was driven by human understanding and experience, and only humans were responsible for the impact of the decisions stemming from such insights. Today, however, machine learning permeates every […]

Abstract: Data-driven systems hold immense potential to positively impact society, but their reliability remains a challenge. Their outputs are often too brittle to changes in their training data, leaving them vulnerable to data poisoning attacks, prone to leaking sensitive information, or susceptible to overfitting. Establishing fundamental principles for designing algorithms that are both stable—to mitigate these […]

Abstract: Controlling language models is key to unlocking their full potential and making them useful for downstream tasks. Successfully deploying these models often requires both task-specific customization and rigorous auditing of their behavior. In this talk, I will begin by introducing a customization method called Prefix-Tuning, which adapts language models by updating only 0.1% of […]

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 […]