CS410.M11 - Neural networks and Genetic Algorithm

This is our implementation for Multi-Objective Optimization for Neural Architecture Search Problem

Requirements

Python >= 3.6.8, PyTorch >= 1.5.1, torchvision >= 0.6.1, pymoo == 0.5.0, nats-bench

Architecture search

To run architecture search:

python main.py --dataset cifar10 --pop_size 200 --n_gens 250 --n_offspring 20

Visualization

To visualize the architectures:

python misc/visualize.py --dataset cifar10 --type graph           

Remember to update your projecty root path before running

IGD Metrics

	CIFAR-10	CIFAR-100	ImageNet-16
IGD	0.004	0.03	0.03

Results

CIFAR-10	CIFAR-100	ImageNet-16-120