The NeurOps repository implements neural operations, as defined in Maile et al. (2022), to enable architectural optimization via structured growing, pruning, and masking during the standard training process. It is currently available in PyTorch, with planned extensions to Jax/Flax and Julia/Flux (as done in NORTH* Neurogenesis).

Previous Post
Next Post
“Equivariance-Aware Architectural Optimization of Neural Networks” accepted at ICLR