Abstract:

The rise of machine learning (ML) has led to an explosion in potential strategies which may be used to learn from data in order to make scientific predictions. For physical scientists who wish to apply ML strategies to a particular domain, the vast number of strategies available has made it difficult to make an a priori assessment of what strategy to adopt. This is further complicated when similar domains have not been previously explored in the literature.

Recently, we worked with Kaggle to design a competition which encouraged data scientists around the world to develop ML models for predicting pairwise nuclear magnetic resonance (NMR) properties for synthetically relevant chemical compounds. Over 3 months, we received 47,800 ML model submissions from 2700 teams in 84 countries, with the top models outperforming our own previously published methods. The success this strategy has cultivated highlights the potential of crowd-sourced ML approaches across a range of scientific domains.

This symposium will introduce the background and main findings of the competition, including the context of computational NMR, the Kaggle platform and presentations from the top performing teams of the competition.

Location: Lecture Theatre 2.41, Fry Building, Bristol BS8 1TH

Schedule:

10:00 - 11:15: Registration

11:15 - 11:50: Craig Butts and Will Gerrard

11:50 - 12:25: Addison Howard and Walter Reade

12:25 - 13:15: Lunch

13:15 - 13:50: Lars Andersen Bratholm

13:50 - 14:25: Brandon Anderson

14:25 - 14:40: Break

14:40 - 15:15: Luka Stojanovic and Milos Popovic

15:15 - 15:50: Youhan Lee and Sunghwan Choi

15:50 - 16:05: Break

16:05 - 16:40: Andres Torrubia

16:40 - 17:15: Devin Wilmott

17:15 - 18:15: Drinks reception

Sponsors:

CHAMPS - Chemistry and Mathematics in Phase Space (CHAMPS) is a 6 year EPSRC-sponsored Programme Grant involving Bristol University, Cardiff University, Imperial College and Leeds University.

JGI - ‘The Jean Golding Institute fosters high quality data-intensive research. We facilitate and strengthen interdisciplinary work, provide data-science expertise, and build a cohesive data science community in an increasingly data-rich world.’