Researcher

Brian McWilliams

Brian McWilliams is a machine-learning researcher whose work spans scalable statistics, rendering, game theory, social language models and generative music and audio. He earned a mathematics PhD from Imperial College London in April 2012, then coauthored work on semi-supervised learning, random-projection optimization, neural denoising and Monte Carlo sampling. EigenGame reframed principal component analysis as a competitive multi-agent game and won ICLR’s 2021 Outstanding Paper Award; he also coauthored EigenGame Unloaded and TwHIN-BERT. His earlier research roles included Disney Research, Twitter and Google Research; at Google DeepMind, he contributed to Lyria in 2023, coauthored speech-generation research behind NotebookLM Audio Overviews in 2024, and coauthored Live Music Models in 2025. As of October 2026, he is an AI research scientist at Google DeepMind, co-leads the Lyria team and is one of Lyria 2’s developers.

Timeline

  1. 2012

    A mathematics PhD

    He earns a PhD in mathematics from Imperial College London in April 2012.

  2. 2013–2016

    Correlated Nystrom and random projections

    In 2013, he coauthors “Correlated random features for fast semi-supervised learning,” which presents Correlated Nystrom Views for regression and classification. In 2016, he coauthors “Scalable Adaptive Stochastic Optimization Using Random Projections” at Disney Research.

  3. 2018–2019

    Denoising and neural sampling

    In 2018, at Disney Research, he coauthors “Denoising with Kernel Prediction and Asymmetric Loss Functions,” a rendered-image denoising method that combines kernel-predicting networks with task-specific modules and asymmetric loss. In 2019, he coauthors Neural Importance Sampling, which uses deep neural networks to generate samples for Monte Carlo integration.

  4. 2021–2022

    EigenGame and TwHIN-BERT

    In 2021, he coauthors EigenGame, which reframes principal component analysis as a competitive multi-agent game and wins ICLR’s Outstanding Paper Award. In 2022, he coauthors EigenGame Unloaded, which proposes an unbiased stochastic update that enables greater parallelism on larger datasets. Also in 2022, he coauthors TwHIN-BERT, a multilingual language model trained on seven billion tweets across more than 100 languages with a social objective based on engagement within Twitter’s heterogeneous information network.

  5. 2023

    Contributing to Lyria

    In 2023, he contributes research to Google DeepMind’s Lyria, its music-generation model.

  6. 2024

    NotebookLM audio generation

    In 2024, he coauthors Google DeepMind speech-generation research underpinning NotebookLM Audio Overviews; the feature turns uploaded documents into dialogue between two AI hosts.

  7. 2025

    Live Music Models

    In 2025, he coauthors Live Music Models with the Lyria Team; it describes systems that generate continuous music in real time with synchronized user control. The paper releases Magenta RealTime as an open-weights model and Lyria RealTime as an API-based model.

  8. As of October 2026

    Lyria 2 at DeepMind

    His earlier research positions include Google Research, Twitter and Disney Research, where he works on machine-learning algorithms for rendering, image and video processing. As of October 2026, he is an AI research scientist at Google DeepMind and co-leads the Lyria team. He is one of the developers of Google DeepMind’s Lyria 2.

Career

  1. NowGoogle DeepMindDeveloper
  2. Until 2019The Walt Disney Company

Sources and credits

Photo: We have no published photo of Brian McWilliams yet. Send us a free-licensed photo.

  1. Projection Based Models for High Dimensional Data spiral.imperial.ac.uk. Checked 22 October 2026.
  2. Correlated random features for fast semi-supervised learning papers.nips.cc. Checked 22 October 2026.
  3. NIPS 2016 Accepted Papers neurips.cc. Checked 22 October 2026.
  4. Denoising with Kernel Prediction and Asymmetric Loss Functions | Disney Research Studios studios.disneyresearch.com. Checked 22 October 2026.
  5. Neural Importance Sampling | Disney Research Studios studios.disneyresearch.com. Checked 22 October 2026.
  6. Game theory as an engine for large-scale data analysis — Google DeepMind deepmind.google. Checked 22 October 2026.
  7. [2102.04152] EigenGame Unloaded: When playing games is better than optimizing arxiv.org. Checked 22 October 2026.
  8. [2209.07562] TwHIN-BERT: A Socially-Enriched Pre-trained Language Model for Multilingual Tweet Representations at Twitter arxiv.org. Checked 22 October 2026.
  9. Transforming the future of music creation — Google DeepMind deepmind.google. Checked 22 October 2026.
  10. Pushing the frontiers of audio generation — Google DeepMind deepmind.google. Checked 22 October 2026.
  11. [2508.04651] Live Music Models arxiv.org. Checked 22 October 2026.
  12. Lyria 3.5 — Google DeepMind deepmind.google. Checked 22 October 2026.
  13. Home sites.google.com. Checked 22 October 2026.