Reproduce “Towards Reasonable Concept Bottleneck Models” 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
Nektarios Kalampalikis, Kavya Gupta, Georgi Vitanov, Isabel Valera
CREAM lets designers constrain how concepts inform one another and outputs in concept bottleneck models, using masked neural networks and an optional, regularized side-channel for information missing from concepts. In CUB, groupwise removal of mutually exclusive concepts left task accuracy nearly unchanged through 55% removal; at dropout p = 0.8, the authors report interpretability with about 10% remaining in this experiment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.