PeopleofSuperintelligence

Researcher

Lucas Theis

Lucas Theis is a machine learning researcher who works on neural compression and on how generative models are evaluated. His PhD at the Max Planck Research School for Neural Information Processing in Tübingen, in Matthias Bethge's lab, produced some of the first deep generative models for images. He joined Twitter in 2016 through the acquisition of Magic Pony Technology, coauthored the SRGAN super-resolution paper in 2016, and was first author of the compressive autoencoders paper in 2017. At Google from 2020, he worked on neural compression, coauthoring a 2022 survey, An Introduction to Neural Data Compression, and the C3 codec in 2023. He is the founder of Mabyduck, a platform for subjective quality evaluation experiments that launched in November 2025.

Now at
Mabyduck
Badges

Timeline

  1. By 2012

    PhD in Bethge's lab

    He earns a BSc in cognitive science at the University of Osnabrück, then does his PhD at the Max Planck Research School for Neural Information Processing in Tübingen, in Matthias Bethge's lab. His doctoral work includes some of the first deep generative models for images. In July 2012, he presents hierarchical models of natural images in a seminar at the Redwood Center for Theoretical Neuroscience.

  2. 2015

    Evaluating generative models

    In November 2015, he posts "A note on the evaluation of generative models" with Aäron van den Oord and Matthias Bethge. The note shows that average log-likelihood, Parzen window estimates and visual fidelity of samples are largely independent of each other for high-dimensional data.

  3. 2016

    Joining Twitter via Magic Pony

    He works at Magic Pony Technology, a London video compression startup, from January 2016, developing early transform coding approaches based on neural networks. In June 2016, he joins Twitter through the acquisition of the company.

  4. 2016–2017

    SRGAN and compressive autoencoders

    In September 2016, he is a coauthor of SRGAN, a generative adversarial network for image super-resolution that infers photo-realistic natural images at 4x upscaling factors. In March 2017, he is first author of "Lossy Image Compression with Compressive Autoencoders" with Wenzhe Shi, Andrew Cunningham and Ferenc Huszár, which trains deep autoencoders competitive with JPEG 2000.

  5. 2020

    Neural compression at Google

    He joins Google in 2020, where he works on compression using neural networks. In June 2020, he and Eirikur Agustsson post "Universally Quantized Neural Compression", which shows that a uniform noise channel can be implemented at test time using universal quantization.

  6. 2022

    Neural compression survey and DiffC

    In February 2022, he posts "An Introduction to Neural Data Compression" with Yibo Yang and Stephan Mandt, a survey of learned data compression. The June 2022 paper "Lossy Compression with Gaussian Diffusion", by him, Tim Salimans, Matthew D. Hoffman and Fabian Mentzer, introduces DiffC, a lossy compression approach based on unconditional diffusion generative models that outperforms HiFiC on ImageNet 64x64.

  7. 2023–2024

    C3 and a universal critic

    In December 2023, he coauthors C3, a neural compression method that fits a small model to each image or video separately and can decode with an order of magnitude lower complexity than neural baselines of similar rate-distortion performance. His ICML 2024 paper "What makes an image realistic?", posted in March 2024, introduces the notion of a universal critic for quantifying realism.

  8. As of October 2026

    Founding Mabyduck

    He leaves Google DeepMind to launch his own startup, Mabyduck. Its self-serve platform for subjective quality evaluation experiments launches in November 2025, and he is its founder.

Career

  1. Since 2025MabyduckFounder
  2. Until 2025Google DeepMind
  3. 2016Magic Pony Technology
  4. From 2016Twitter

Education

  1. University of OsnabrückBSc
  2. Max Planck Research School for Neural Information ProcessingPhD

From the biography.

  1. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  2. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Google. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. Lucas Theis theis.io. Checked 21 October 2026.
  2. Speakers - NASIT 2023 - University of Pennsylvania nasit.seas.upenn.edu. Checked 21 October 2026.
  3. Lucas Theis July 2012 - Redwood Center for Theoretical Neuroscience redwood.berkeley.edu. Checked 21 October 2026.
  4. A note on the evaluation of generative models arxiv.org. Checked 21 October 2026.
  5. Lucas Theis - LinkedIn, Twitter, Facebook me.sh. Checked 21 October 2026.
  6. Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network arxiv.org. Checked 21 October 2026.
  7. Lossy Image Compression with Compressive Autoencoders arxiv.org. Checked 21 October 2026.
  8. Universally Quantized Neural Compression arxiv.org. Checked 21 October 2026.
  9. An Introduction to Neural Data Compression arxiv.org. Checked 21 October 2026.
  10. Lossy Compression with Gaussian Diffusion arxiv.org. Checked 21 October 2026.
  11. C3: High-performance and low-complexity neural compression from a single image or video arxiv.org. Checked 21 October 2026.
  12. What makes an image realistic? arxiv.org. Checked 21 October 2026.
  13. Simplifying perceptual quality evaluation - Mabyduck mabyduck.com. Checked 21 October 2026.