Bayesian Optimization of Neural Architecture

We implemented Neural Architecture Search with Bayesian Optimisation and Optimal Transport as course project during our exchange period.

It was a cherishing experience to take Bayesian Methods in Machine Learning held by professor Roman Garnett during my exchange period in Washington University in St. Louis. As our course project, my teammates and I managed to finalize the challenging implementation of this paper within 2 weeks and used it to optimize neural network architecture to classify cifar-10, whose average accuracy was tuned from 0.5 to 0.82. It’s very interesting and we expect more researches on AutoML.

You can view our code, report and presentation slides here. [code] [report] [slides]