Understanding polysemanticity in neural networks through coding theory

Simon C. Marshall, Jan H. Kirchner

Despite substantial efforts, neural network interpretability remains an elusive goal, with previous research failing to provide succinct explanations of most single neurons' impact…