[ICLR'23] Trainability Preserving Neural Pruning (PyTorch)

"Module pytrees" that cleanly handle parameter trainability and transformations for JAX models.

Simulations and analysis showing that gradient loss in noisy U(1)-equivariant quantum neural networks is governed by readout-visible sector coherence. Density-matrix simulations, regression analysis, and reproducibility code for a study of noise-induced gradient degradation in equivariant brickwork QNNs.

A representation-theoretic trajectory diagnostic for quantum neural networks. The symmetry-organised complexity index measures how a QNN distributes expressive capacity across the multiplicity structure of a target symmetry, rather than how much of Hilbert space it visits.