About. I’m a researcher with wide-ranging interest across AI and machine learning. I am based at Microsoft Research working with the Game Intelligence team (Cambridge), Machine Learning Group (Beijing), & AI Frontiers (NYC).
Previous. I completed my PhD at the University of Cambridge, spending time at the Alan Turing Institute, and then did a postdoc at Tsinghua University. In my previous life I qualified as a Chartered Accountant with Ernst & Young (EY), building forecasting and valuation models for clients across London finance. I also spent one year in Taiwan studying 中文.
Research. I have deep expertise in scaling laws, imitation learning, world modeling, generative modeling, and uncertainty estimation. My PhD centered around Bayesian deep learning (e.g. ensembling and priors), I then moved focus to applying ideas from generative modeling to embodied settings (e.g. using diffusion to imitate humans and game environments). My work (and dataset) on behavioral cloning for video games has been widely shared. Recently I have helped understand the science of scaling in AI, such as reconciling conflicting scaling law coefficients.
My research aims to build and understand AI systems that are scalable, robust and human-like. My philosophy incorporates several elements. 1) I adopt a probabilistic view of neural networks to help understand today’s algorithms and design those of tomorrow. 2) I believe that the agent-environment framework, with sequential decisions and interactive learning, is the correct setting to be studying to make long-term progress in AI. 3) My research places equal weighting on theory and empirics.
Get in Touch. If you’d like to chat about research, collaborating together, or other opportunities, reach me via email at $x$@microsoft.com, where, $x=\text{v-timpearce}$. I’m more active on Twitter than here, so for up-to-date news, follow me.
News
- October 2024. Two papers accepted to NeurIPS! Diffusion for World Modeling: Visual Details Matter in Atari, C-GAIL: Stabilizing Generative Adversarial Imitation with Control Theory.
- August 2024. New preprint out Reconciling Kaplan and Chinchilla Scaling Laws.
- May 2024. New preprint out 👏 Diffusion for World Modeling: Visual Details Matter in Atari.
- May 2024. Presenting two papers at ICLR workshop generative AI for decision making.
- March 2024. Wrote a blog post tracking the GPT series of models here.
- July 2023. Giving a talk at QMU’s Game AI group. 👾
- June 2023. Co-organizing Summer School on AI and Games.
- June 2023. Traveling to Manchester Unveristy to give a keynote at Advances in Data Science and AI Conference.
- May 2023. Heading to Rwanda for ICLR! ✈️
- Jan 2023. 1x ICLR paper accepted. 💫
- Jan 2023. Giving a talk at Tsinghua TSAIL group on diffusion models for imitation learning.
- Dec 2022. Attending NeurIPS to present my paper on censored quantile regression neural networks.
- July 2022. My paper on behavioural cloning for counter-strike won best paper at IEEE CoG.
- May 2022. I will be joining Microsoft Research as an AI researcher.
Misc Resources
- I sometimes make short videos summarising research papers I find interesting, e.g. a robotic system that plays table tennis with RL, or explaining how AlphaCode works
- I used MusicGen and other AI tools to produce endless lofi music – explanation tweet here.
- I demonstrated how classifier-free guidance for diffusion works using a minimal repo.