Research

Graph theory and learning in games: papers, talks and open projects.

My research sits between two worlds. From the combinatorics side, I work on extremal and probabilistic questions about graphs and hypergraphs (colourings, Ramsey numbers, embeddings, sparse regularity). From the game theory side, I study what learning dynamics actually do when several agents use them at once: when they reach equilibria, when they collude, and when they fail to discard strategies a rational player never would.

Papers

If you want an overview of my research so far, have a look at the introduction to my PhD Thesis: Colours, Giants, and Games.

No matching items

Talks

A selection.

No matching items

Future projects

I welcome collaboration with anyone who would like to work with me on one (or more) of the following. These are only the projects for which I HAVE a clear research path in mind, with specific implementations, experiments, and analysis to do, I just didn’t have the time to implement them yet. Please get in touch with me even just to chat about them, or if you’re interested in collaborating.

Can RL learn Combinatorial Games?

Reinforcement Learning (RL) is really good at games, but is possibly really bad at Graph Theory (although initial results were quite optimistic [1, 2]). Would RL be good at Combinatorial Games? I am particularly interested to explore this about games on graphs. In particular the Balance Game or some Maker Breaker games.

What can RL do in Graph Theory?

There are some results that seem to indicate that RL is not good at Graph Theoretical problems [3, 4]. However, RL is clearly good at solving certain Graph-theoretical problems, and initial results were quite optimistic [1, 2]. What is the reason behind this varying behaviour?

Sparsity? Quantization? What about both?

Two well-known memory-reduction methods in Neural Networks are Sparsity and Quantization. These two methods are not orthogonal. However I do believe that there is a way of improving the current SotA to produce a lighter version of NN using the interplay between the two. The literature in the area seems to go towards a natural direction that hasn’t been developed, in my opinion, to the fullest.

Let’s make EU healthier

This is a project that has a wider breadth and is not only linked to research. If you’re passionate about nutrition, hardcore computer science, and want to make a tangible impact, do get in touch.

Theses and projects

Service and awards

  • Organiser
    • PCC (Main organiser), Apr 2023, University of London and BCC. (National conference for postgraduate students).
    • LSE PhD CGO Seminar (Main Organiser), 2022/23.
  • Grants and invitations
    • DMV Invited Speaker. Ilmenau. 2023.
    • LMS Computer Science Small Grant (700£). London Mathematical Society. Funded visit to Berlin. 2022.
    • LSE Contribution Award. Dept of Mathematics. LSE. 2021.