4 problems

A Hopfield network is a neural network of $N$ neurons whose stored patterns are stable configurations of its dynamics. Linear capacity conjecture. The number of such patterns for a…

Rician-training benefit conjecture. Rician training is more beneficial for textsc{TurboAE-TI} because of its sub-optima problem.

Rician-training gradient-perturbation conjecture. Rician training helps textsc{TurboAE-TI} escape local sub-optima by injecting gradient perturbations during encoder training.

Sparse-coding conjecture. Sensory systems may encode stimuli via sparse approximation. This conjecture connects probabilistic accounts of perception with the theoretical benefits o…