10.5258/SOTON/AI3SD0171
YUAN, YINGFANG
YINGFANG
YUAN
https://orcid.org/0000-0002-8925-9267
Heriot-Watt University
AI3SD Video: Hyperparameter Optimisation for Graph Neural Networks
University of Southampton
2021
Video
AI, AI3SD Event, Artificial Intelligence, Chemical Space, Chemistry, Machine Learning, ML, Neural Networks
Frey, Jeremy
Jeremy
Frey
https://orcid.org/0000-0003-0842-4302
University of Southampton
Kanza, Samantha
Samantha
Kanza
https://orcid.org/0000-0002-4831-9489
University of Southampton
Niranjan, Mahesan
Mahesan
Niranjan
https://orcid.org/0000-0001-7021-140X
University of Southampton
2021
en
https://www.youtube.com/watch?v=gfkQ3-G2G6M
mp4
Creative Commons Attribution 4.0 International
Traditional deep learning has made significant progress on various problems, from computer vision to natural language processing. For graph problems, there are still many challenges. Graph neural networks (GNNs) have been proposed for a wide range of learning tasks in the graph domain. In particular, in recent years, an increasing number of GNN models were applied to model molecular graphs and predict the properties of the corresponding molecules. However, a direct impediment to achieve good performance with the lower computational cost is to select appropriate hyperparameters. Meanwhile, many molecular datasets are far smaller than many other datasets in typical deep learning applications. Most hyperparameter optimization (HPO) methods for deep learning have not been explored in terms of their efficiencies on such small datasets in the molecular domain. We conducted theoretical analyses for popular HPO methods (random search, TPE, and CMA-ES) and proposed a genetic algorithm with hierarchical evaluation strategy and tree-structured mutation for HPO. Finally, we believe that our work will motivate further research to GNNs as applied to molecular machine learning problems and facilitate scientific discovery.
Engineering and Physical Sciences Research Council
https://doi.org/10.13039/501100000266
EP/S000356/1
Artificial and Augmented Intelligence for Automated Scientific Discovery