Seeing is Believing: Brain-Inspired Modular Training for Mechanistic Interpretability

May 04, 2023 · Entered Twilight · 🏛 arXiv.org

Repo contents: README.md, incontext_3.4.ipynb, mnist_3.5.ipynb, modular_addition_3.3.ipynb, permutation_S4_3.3.ipynb, requirements.txt, symbolic_formulas_3.1.ipynb, two_moon_3.2.ipynb

Authors Ziming Liu, Eric Gan, Max Tegmark arXiv ID 2305.08746 Category cs.NE: Neural & Evolutionary Cross-listed cond-mat.dis-nn, cs.AI, cs.LG, math.RT, q-bio.NC Citations 50 Venue arXiv.org Repository https://github.com/KindXiaoming/BIMT ⭐ 175 Last Checked 1 month ago

Abstract

We introduce Brain-Inspired Modular Training (BIMT), a method for making neural networks more modular and interpretable. Inspired by brains, BIMT embeds neurons in a geometric space and augments the loss function with a cost proportional to the length of each neuron connection. We demonstrate that BIMT discovers useful modular neural networks for many simple tasks, revealing compositional structures in symbolic formulas, interpretable decision boundaries and features for classification, and mathematical structure in algorithmic datasets. The ability to directly see modules with the naked eye can complement current mechanistic interpretability strategies such as probes, interventions or staring at all weights.