3 citations · 3 across the 5 of their papers we have counts for
5 papers
Turn-Averaged SAEs for Feature Discovery and Long-Context Attribution
Kevin Der, Harish Kamath, Ben Thompson
Sparse autoencoders (SAEs) have become a useful tool for extracting interpretable features in language models. However, standard SAE architectures operate on individual token activ…
Distillation Robustifies Unlearning
Bruce W. Lee, Addie Foote, Alex Infanger +6
Current LLM unlearning methods are not robust. A few steps of finetuning can revert their effects. We begin by showing that this is true even for an idealized form of unlearning: t…
Hierarchical Classification System for Breast Cancer Specimen Report (HCSBC) -- an end-to-end model for characterizing severity and diagnosis
Thiago Santos, Harish Kamath, Christopher R. McAdams +8
Automated classification of cancer pathology reports can extract information from unstructured reports and categorize each report into structured diagnosis and severity categories.…
Direct Inversion: Optimization-Free Text-Driven Real Image Editing with Diffusion Models
Adham Elarabawy, Harish Kamath, Samuel Denton
With the rise of large, publicly-available text-to-image diffusion models, text-guided real image editing has garnered much research attention recently. Existing methods tend to ei…
Model-Advantage and Value-Aware Models for Model-Based Reinforcement Learning: Bridging the Gap in Theory and Practice
Nirbhay Modhe, Harish Kamath, Dhruv Batra +1
This work shows that value-aware model learning, known for its numerous theoretical benefits, is also practically viable for solving challenging continuous control tasks in prevale…