Skip to main content Skip to navigation

George Watkins

Current Research

I am a PhD student in the Mathematics for Real-World Systems CDT at the University of Warwick. My research interests are in 𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲, in particular 𝗗𝗲𝗲𝗽 𝗥𝗲𝗶𝗻𝗳𝗼𝗿𝗰𝗲𝗺𝗲𝗻𝘁 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴.

My current research focuses on 𝗥𝗲𝗹𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗥𝗟, exploring new methods for learning in environments in which the 𝘀𝘁𝗮𝘁𝗲𝘀 𝗮𝗿𝗲 𝗿𝗲𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗲𝗱 𝗮𝘀 𝗴𝗿𝗮𝗽𝗵𝘀.

Most recently I have been investigating how Relational RL can be used to generate heuristic methods for 𝗖𝗼𝗺𝗯𝗶𝗻𝗮𝘁𝗼𝗿𝗶𝗮𝗹 𝗢𝗽𝘁𝗶𝗺𝗶𝘀𝗮𝘁𝗶𝗼𝗻 𝗽𝗿𝗼𝗯𝗹𝗲𝗺𝘀. I have explored the application of Reinforcement Learning to the Graph Colouring problem and the Wedding Seating Plan problem, and have demonstrated that such methods have the ability to generalise effectively to new, previously unseen, problem parameterisations.

Education

  • 2019-present: PhD in Reinforcement Learning, exploring memory in partially observable environments
  • 2018-2019: MSc in Mathematics for Real-World Systems, University of Warwick (Distinction)
  • 2011-2012: PGCE, Institute of Education
  • 2004-2008: MSci in Mathematics, Imperial College (First Class Honours)

Teaching

  • 2019-2020: Interdisciplinary Approaches to Machine Learning, University of Warwick
  • 2018-2020: Quantitative Analysis for Management, Warwick Business School
  • 2014-2016: Head of Year, King's College London Mathematics School (years 12-13)
  • 2011-2014: Teacher of Mathematics, St. Marylebone School (years 7-13)

Talks & Presentations

  • 2022
    • Alan Turing Institute Networking Event
      • Using Reinforcement Learning to learn a heuristic for the Graph Colouring problem
    • Mathematics for Real World Systems Annual Conference
      • Colouring: Not as Therapeutic as Advertised
  • 2020
    • Machine Learning, Reinforcement Learning and Bayesian Optimisation Reading Group
      • Reinforcement Learning algorithms
      • Policy Gradient methods
      • Variational Auto-Encoders
  • 2019
    • Machine Learning, Reinforcement Learning and Bayesian Optimisation Reading Group
      • Loss functions in ML
      • Optimization Functions in ML
    • University of Warwick Data Science Reading Group
      • Deriving Policy Gradient methods

Other contributions

  • Alan Turing Institute Reinforcement Learning Study Group

In February 2021 I was fortunate to have the opportunity to lead a team in the 2-week RL Data Study Group. In collaboration with the Defence Science and Technology Laboratory (Dstl) we explored whether RL can provide solutions which are adaptable to changes in the configuration of the environment.

The final report can be found here: https://doi.org/10.5281/zenodo.5121558Link opens in a new window

  • Founder of Warwick University's Machine Learning, Reinforcement Learning and Bayesian Optimisation Reading Group

Talks covered key ideas from each of the fields, as well as providing an opportunity to discuss recent research and collaborate on projects.

Information about the reading group can be found here: https://warwick.ac.uk/fac/sci/mathsys/news/readinggroups/machinelearningrg/Link opens in a new window

  • Contributor to PyTorch Geometric and NetworkX libraries

Awards & Achievements

  • Best presentation - Mathsys Annual Conference, 2022
  • Top of Cohort - Mathematics for Real World Systems Master's Programme, 2018-19

George Watkins