Neural networks learn to solve hard problems by converging to stable states

Original: Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning

Source: arxiv.org ↗