Reproduce “Neural Basis Models for Interpretability” 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

Filip Radenovic, Abhimanyu Dubey, Dhruv Mahajan

Neural Basis Models (NBMs) build additive feature explanations by sharing a network that produces basis functions, then combining them with feature-specific weights. This design retains independent feature contributions while reducing size and inference cost relative to Neural Additive Models (NAMs). On CoverType’s 54 features, NBM uses 70K parameters and processes 530K inputs per second, with similar top-1 accuracy: 0.7369 versus 0.7359 for NAM.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.