Human-Centered Learning Mechanics: A Dynamical Framework for Entropy-Regulated Representation Learning

In three linesHCLM theoretical framework for entropy-regularized learning in open dynamical systems. Introduces effective information force concept and characterizes degenerate entropy regimes. Shows geometric entropy surrogates (variance, log-determinant covariance) produce stronger, more stable gradients than softmax-normalized entropy.

Summary generated by Claude — human-verified