Back to projects

Evolutionary Neural Network Optimization

View resource →

  • PyTorch
  • evolutionary algorithms
  • optimization
  • AutoML

What it is

An experiment in using evolutionary algorithms to optimize neural-network training hyperparameters (e.g. learning rate, momentum).

It implements both single-objective and multi-objective evolution — the multi-objective variant searches for a Pareto front trading off final accuracy against time-to-target-accuracy. Candidates are evaluated with partial training, then the best is fully trained and logged. Includes baseline comparisons and plotting utilities across multiple epoch budgets.

Stack

Python · PyTorch · custom evolutionary/genetic operators · Pareto-front selection

Links