AutumnTHE OLYMPIAD COLLECTION
Menu
RESEARCHED WITH AUTUMN
HM

Hartmut Maennel

Staff Research Scientist at Google DeepMind; mathematician (PhD, University of Bonn)
Google DeepMind · Zurich, Switzerland · engineer

Mathematician and ML researcher. IMO gold medalist for Germany (1979, 1981); PhD mathematics, University of Bonn (1992); Member, Institute for Advanced Study Princeton (1992-93). Career spans software engineering at Microsoft and Google, a period as Senior Vice President, Research at the quantitative investment manager Winton Group (2011-2017), and current work as Staff Research Scientist at Google DeepMind Zurich on machine learning and quantum-chemistry/molecular simulation.

Hartmut Maennel is a mathematician and machine-learning researcher. He represented Germany at the International Mathematical Olympiad in 1978, 1979 and 1981, winning gold medals in 1979 and 1981. He completed a PhD in mathematics at the University of Bonn (1992) and was a Member in the School of Mathematics at the Institute for Advanced Study, Princeton (9/1992-4/1993). He subsequently held roles at Microsoft, Google (software/research engineering), and as Senior Vice President, Research at the quantitative investment manager Winton Group (Dec 2011-Aug 2017). He is now a Staff Research Scientist at Google DeepMind in Zurich, publishing on machine learning and molecular quantum properties.

Details
LocationZurich, Switzerland
Company sitedeepmind.com
UniversityUniversity of Bonn
Machine learningMathematicsNeural network theoryMolecular simulation / quantum chemistry MLStatisticsComputational quantum chemistry
Notes
  • Quant employment verification: CONFIRMED - an actual paid research role at a systematic quantitative investment manager. LinkedIn enrichment shows Hartmut Maennel as Senior Vice President, Research at Winton Group from December 2011 to August 2017 (index-only claim, not independent confirmation). Winton's own site corroborates the employer's classification as a quantitative investment manager: it 'research[es], design[s] and trade[s] systematic investment strategies' across thousands of exchange-traded and OTC instruments and states it is a leader in trend following. So the employer is genuinely quantitative, and the role is a substantive research position (not a student program). The person's identity is supported independently (IMO, IAS, ORCID). Caveat: the Winton role itself rests on the LinkedIn enrichment index, not on a primary Winton document naming him.Sep 17, 2026
  • Index-only caveat: the employment timeline other than the IAS membership comes from a LinkedIn enrichment index (ClickHouse-backed), not from primary Winton/Google/Microsoft documents naming him; it should be treated as an unverified index claim rather than independent confirmation.Sep 17, 2026
  • Olympiad record: represented Germany at the IMO in 1978 (21 points), 1979 and 1981. Won gold in 1979 (39/42, rank 5, 97.6 percentile) and gold in 1981 (42/42, rank 1, 100 percentile). Note the caller brief listed only 1979 and 1981; the official record also shows a 1978 participation.
  • Current role: Research Scientist at Google (Switzerland), Zurich, in Google DeepMind (ORCID employment entry, self-asserted; Google Scholar confirms affiliation 'Google DeepMind', verified google.com email). The caller LinkedIn lead still shows the same employer/title, cross-checking the current role.
  • Employer classification check: Winton Group (winton.com) is a real systematic/quantitative investment manager ('We research, design and trade systematic investment strategies ... provide diversification ... leader in trend following'), a hedge-fund/quant-investment firm rather than a company merely containing 'quant' in its name.
  • Education: PhD in mathematics from the University of Bonn, 1992, per the IAS scholar record; LinkedIn enrichment also lists Mathematics at the University of Cambridge. Both are degrees/study, not employment.
  • Research output: publishes machine-learning theory and quantum-chemistry/molecular-simulation methods - e.g. 'Gradient Descent Quantizes ReLU Network Features' (arXiv 1803.08367, 2018), 'What Do Neural Networks Learn When Trained With Random Labels' (NeurIPS 2020), 'Deep Learning Through the Lens of Example Difficulty' (NeurIPS 2021), and 'Complete and Efficient Covariants for 3D Point Configurations' (arXiv 2409.02730, 2024). Google Scholar lists 1180 citations, h-index 14.
  • Post-doctoral research: Member in the School of Mathematics at the Institute for Advanced Study, Princeton, 9/1992-4/1993 (listed as a Past Member, Mathematics).Sep 1992
  • Patents assigned to Google LLC list him as inventor, e.g. US 12675687 'Training neural networks with reinitialization' (granted 2026-07-07) and application 20260024019 on rotationally invariant/covariant descriptors of point configurations.
  • No GitHub, X/Twitter, blog or personal website was located; the public footprint is researcher profiles (ORCID, Google Scholar, DBLP, patents) plus LinkedIn. No public email is published beyond tool-provided and company-domain hints - a RocketReach listing shows only pattern-guessed gmail/googlemail/deepmind addresses (not saved as verified contact).

Competition record

Germany · IMO

Related collections

Explore the collection