Writing

  1. A solvable model of feature learning in a nonlinear networkA statistical mechanics account of feature learning in a Bayesian network with two trainable layersA statistical mechanics account of how a Bayesian network changes its hidden weight distribution to learn one useful direction.
  2. Mathematical properties of Gaussian process neural codeExcursion geometry, spectral structure, and coding errorA mathematical field note on Gaussian process neural codes, excursion geometry, spectral structure, and coding error.