AI, ML and Friends is a weekly seminar series within the School of Computing on Artificial Intelligence, Machine Learning, and related topics. We are open to attendees and presenters external to the school. Please sign up to the mailing list to receive weekly announcements including zoom details, and email the seminar organiser to schedule a talk.

Upcoming Seminars

05 March 2026, 11:00#

Partially Observable MDPs in Aerospace Applications#

Speaker: Prof Nicolas Drougard

Abstract: The Partially Observable Markov Decision Process (POMDP) model gave us a way to effectively solve sequential decision marking problems for aerospace applications. First, problems with known models could be simplified using possibility theory to make computation times more feasible. For problems with unknown models, data augmentation methods based on the symmetry properties of the systems were proposed. Finally, the development of an offline reinforcement learning algorithm enabled the control of human-machine interaction for a robot teleoperation problem, based on real past data trajectories, including features derived from physiological data. We will conclude this presentation with ongoing work on the control of drones and life support systems for space exploration.

Bio: After a PhD thesis in Planning under Uncertainty at Onera defended in 2015, Nicolas Drougard worked as a postdoc in Artificial Intelligence for Humain-Machine Interaction and Brain Computer Interfaces. From 2019 he is associate professor in AI for driving autonomous systems and space vehicles at ISAE-SUPAERO (Toulouse, France) where he has supervised three defended PhD theses, and is currently supervising two PhD students. Nicolas is working in the fields of Planning, (Offline) Reinforcement Learning and (PO)MDPs, as well as Supervised Learning and Computer Vision, with applications in Aerospace (e.g. UAVs and Bio-regenerative Life Support Systems), Robotics, Agriculture and Human-Machine Interaction.

Where: Skaidrite Darius Building, N101

20 March 2026, 11:00#

What matters for code evolution?#

Speaker: Yonatan Gideoni

Abstract: Recent works like AlphaEvolve demonstrated LLM-based search pipelines solving various problems by finding computer programs for them. Most code evolution pipelines consist of many design choices which are not thoroughly ablated, making it difficult to understand their importance. Testing two simple baselines, we find that much simpler methods can match or exceed purpose-built code evolution pipelines across three domains, each having different constraints. In this talk I will discuss why simple baselines do so well, shortcomings in how code evolution is evaluated and applied, and broader implications for the field.

Bio: Yonatan is a DPhil student at Oxford interested in developing fundamental machine learning methods inspired by human behaviour. This includes understanding when a behaviour’s emergence is limited by a method’s design or an environment, with recent applications in AI assisted search and multimodality. Previously, Yonatan obtained his master’s in computer science from the University of Cambridge and worked on maps for autonomous vehicles at Mobileye. His PhD is funded by the AIMS CDT and a Rhodes scholarship.