


Professor of Computer Science, UCL


Check out recent work on learning to discover.
Large Discovery Models: learning where to search next An LLM proposes, a Bayesian surrogate scores uncertainty, and the loop iterates — cutting validation error by 2.4x, binding energy by 18%, and boosting molecular objectives by 60%+ across programs, proteins, and molecules.

Large Discovery Models: learning where to search next An LLM proposes, a Bayesian surrogate scores uncertainty, and the loop iterates — cutting validation error by 2.4x, binding energy by 18%, and boosting molecular objectives by 60%+ across programs, proteins, and molecules.

Fine-tuning LLM Agents without Fine-tuning LLMs Catchy title and very cool memory technique to improve deep research agents. Great for continuous, real-time learning without gradient updates. Here are my notes:

I guess we will have a lot to talk about #ChatGPT in our upcoming #wsdm2023 tutorial “proactive conversational agents.” I don’t have any concern/complaint/resistance for this intelligent new being. I look forward to my own evolution via interactions w/ this true AI.
Home, Sweet Home. Will be presenting a WSDM'23 Tutorial on "proactive conversational agents" w/ @lizi_liao and @chirag_shah in Singapore, Sweet Singapore.

Congratulations to UKRI Board members Professor Anthony Finkelstein (@profserious), who has received a knighthood, and @JohnFingleton1, who has received a CBE. #NewYearHonours

Check out our multiagent reinforcement learning library together with Shanghai Jiaotong: reddit.com/r/reinforcementle…

So proud! Our new hire @HarrieOos wins SIGIR best paper with a single-authored contribution, and former student @seawan has an honerable mention for the Test-of-time award for his paper on Portfolio theory for IR!! #SIGIR2021

Our weird behavior during the pandemic is messing with AI models technologyreview.com/2020/05…

Beeswax Buys MediaGamma to Shore Up Programmatic Offering adweek.com/programmatic/bees… via @Adweek


An exciting demonstration of curiosity driven learning in #minecraft @Project_Malmo
And what about a mountain car embedded in a 3D #minecraft world, without ANY reward? Well, curiosity-driven learning also gets the cart up the hill here! (paper to come soon, and congrats to Remy Portelas) youtube.com/watch?v=CTuHll7G… @katjahofmann

We provide a new theoretical analysis to understand how they happen, and what we can do about them: arxiv.org/abs/1902.10730

"Artificially Intelligent Players Invent Nonverbal 'Languages' to Win Card Games": My piece for @IEEESpectrum on bridge and Hanabi research by @seawan @ucl @j_foerst @UniofOxford @DeepMindAI spectrum.ieee.org/tech-talk/…

See our recent work on AI playing the Bridge Card game and how AI agents can learn communication language secretly.
"Artificially Intelligent Players Invent Nonverbal 'Languages' to Win Card Games": My piece for @IEEESpectrum on bridge and Hanabi research by @seawan @ucl @j_foerst @UniofOxford @DeepMindAI spectrum.ieee.org/tech-talk/…

Today @UCL launches #MadeAtUCL to showcase our ground-breaking research and discoveries changing people's lives. Know some amazing UCL research that deserves the spotlight? Submit it here: https://t.co/7sjZMl3r17 #MadeAtUni