ES and GA on MNIST

A set of neural networks implemented in tensorflow for testing out evolutionary strategies [1] and a simple genetic algorithm on the MNIST dataset

Scripts

Script	Discription
conv_bp.py, mlp_bp.py	covnet and mlp trained with SGD
mlp_es.py	simple mlps trained with ES, can train any arbitrary architecture
mlp_ga.py	simple mlps trained with GA

References

[1] Salimans, Tim, et al. "Evolution strategies as a scalable alternative to reinforcement learning." arXiv preprint arXiv:1703.03864 (2017).