Abstract

Energy landscapes: from molecules to machine learning

The potential energy landscape provides a conceptual and computational framework for investigating structure, dynamics and thermodynamics in atomic and molecular science. This talk will highlight how new approaches for global optimisation, enhanced sampling of systems exhibiting broken ergodicity, and rare event dynamics can provide new insight into the solution landscape for neural networks. The key aim is to explain how the energy landscape perspective can unify our understanding of apparently disparate systems. A range of applications will be presented including recent results for machine learning landscapes.

Selected Publications: Perspective: New Insights From Loss Function Landscapes of Neural Networks. Machine Learning: Science and Technology, in press, 2020 Machine learning landscapes and predictions for patient outcomes. R Soc Open Sci 4, 170175, 2017. Perspective: Energy Landscapes for Machine Learning, PCCP , 19, 12585-12603, 2017. Feature Article: Exploring Biomolecular Energy Landscapes, Chem. Commun., 53, 6974, 2017 Machine learning prediction for classification of outcomes in local minimisation. Chemical Physics Letters 667, 158, 2017 Exploring Energy Landscapes. Ann. Rev. Phys. Chem., 69, 401-425, 2017 Energy Landscapes: Some New Horizons, Curr. Op. Struct. Biol., 20, 3, 2010. Energy Landscapes, Cambridge University Press, Cambridge, 2003