Reproduce “DeepBase: Deep Inspection of Neural Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Thibault Sellam, Kevin Lin, Ian Yiran Huang, Yiru Chen, Michelle Yang, Carl Vondrick, Eugene Wu
DeepBase provides a declarative system for inspecting hidden-unit behavior against user-specified hypotheses, such as SQL grammar rules or part-of-speech tags. It extracts behaviors and labels, compares them with statistical measures, and uses caching, streaming, convergence-based stopping, and parallel computation. Across tested SQL experiments, it averaged 72× faster than PyBase, but alignment scores do not prove meaning or causality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.