Reproduce “Neural Ideals and Neural Codes: An Algebraic Framework for Neural Network Classification and Feature Interpretation” 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
Venkata Subbaiah Yerrapati, Rahul Dixit, Ajay Kumar Shukla
The paper presents a method for relating hidden neuron activity to classification by binarizing neuron states into neural codes and constructing class-specific Boolean-polynomial constraints from forbidden codes. It also averages active and inactive MNIST inputs and compares them with signed output weights. The XOR example recovers a simple class rule; MNIST visualizations remain qualitative and do not establish explanations or improved accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.