Alistair Benford presenting our NeurIPS paper on runtime analysis of co-evolutionary algorithms applied to combinatorial games. Although the strategy space is exponentially large, the algorithm learns to play optimally in expected quadratic time. openreview.net/forum?id=wWS...

Runtime analysis of evolutionary algorithms counts fitness function evaluations to optimum. We introduce cost models where eval costs differ among search points, allowing cost adaptive optimisation: find optimum by exploring cheaper parts of search space.

link.springer.com/article/10.1...

Learning in games usually assumes small action spaces. This afternoon at #AAAI2025 we give an oral presentation showing that the PDCoEA co-evolutionary algorithm finds the Nash Equilibrium of the game below (2^n actions) in expected poly(n) time. Joint work with Shishen Lin.

Reinforcement learning gathering a crowd at #AAAI

Arrived in Philadelphia for the AAAI conference. We will present a runtime analysis of the PDCoEA (a coevolutionary algorithm) showing logarithmic (wrt number of strategies) expected runtime to find a Nash equilibrium in the game we studied.

COST Action CA22137 ROAR-NET organizes two exciting events this year, the Training School and Code Fest, and has calls open for Short-Term Scientific Missisions (STSMs) and Young Researcher and Innovator Conference Grants roar-net.eu