Research

Compute-Efficient Neuroevolution for Deep Architectures

Research on compute-efficient neural architecture optimization through weight inheritance and genetic mutation.

Validated1 Aug 2023 — 31 Dec 2024 · 1 yr 4 mo

PyTorchPythonGenetic Algorithms

Investigation into weight preservation mechanics in evolving deep convolutional and transformer backbones on resource-constrained hardware.

2024-04-12 · confidence high
Parent weight inheritance eliminates up to 40% of standard cold-start training epochs during evolutionary topology mutation.

Connected work

Projects

Loading content…