Evolving a Neural Network using Genetic Algorithm

Under the supervision of Dr. Anxiao Jiang, I implemented the evolutionary based genetic algorithm as specified in Hinz et al.‘s paper. It’s currently limited to only simple CNNs and uses the Keras library to build, train and validate. Right now, the algorithm is able to achieve 67.88% accuracy on the CIFAR10 dataset, with random seeds, 5 iterations and 10 networks each.

Rose Lin
Rose Lin
Data Scientist @ State Farm

Data Wrangler & Storyteller.

Related