PeopleofSuperintelligence

Researcher

Michael Harrison

Michael Harrison is a mathematician turned AI researcher whose work spans geometric topology, language models, and synthetic data. He earned a mathematics PhD at Pennsylvania State University in 2017, completed postdoctoral work at Lehigh, MSRI, Carnegie Mellon, and the Institute for Advanced Study, and wrote a 2019 paper on totally nonparallel immersions. After his mathematics work, he joined Microsoft Research AI Frontiers in 2024 and worked on Phi-4, whose December 2024 report described a 14-billion-parameter model that used synthetic data throughout training and surpassed its GPT-4 teacher on STEM-focused question answering. The Fara-7B team thanked him for valuable help, discussions, and continued support in November 2025; he coauthored the Phi-4-reasoning-vision-15B report in March 2026 and the Rho report in September 2026. As of October 2026, he is a Member of Technical Staff at Microsoft Research AI Frontiers, working with a team training a vision-language model and focusing on data quality and synthetic data generation.

Now at
Microsoft
Works
32
Citations
217
h-index
8
Badges

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

  1. 2017

    Mathematics PhD at Penn State

    He completes a PhD in mathematics at Pennsylvania State University in 2017, advised by Sergei Tabachnikov.

  2. 2019

    Totally nonparallel immersions

    In 2019, he submits a paper proving that every n-manifold admits a totally nonparallel immersion into a space of dimension 4n−1.

  3. By 2024

    Postdoctoral work in mathematics

    After his 2017 mathematics PhD, he completes postdoctoral work at Lehigh University, MSRI at Berkeley, Carnegie Mellon, and the Institute for Advanced Study at Princeton.

  4. 2024

    Phi-4 at AI Frontiers

    After his mathematics work, he joins AI Frontiers in 2024 and works on the release of Phi-4. In December 2024, he coauthors its technical report, which describes a 14-billion-parameter model that uses synthetic data throughout training and surpasses its GPT-4 teacher on STEM-focused question answering.

  5. November 2025

    Fara-7B project support

    In November 2025, the Fara-7B team thanks him for valuable help, insightful discussions, and continued support on its 7-billion-parameter agentic small language model for computer use.

  6. March 2026

    Phi-4 reasoning vision

    In March 2026, he coauthors the Phi-4-reasoning-vision-15B technical report. The report presents a 15-billion-parameter open-weight multimodal model for vision-language tasks, math and science reasoning, and understanding computer and mobile screens.

  7. September 2026

    Rho manipulation models

    In September 2026, he coauthors the Rho report, which introduces open-weights VLA models for bimanual manipulation designed for data-light task adaptation.

  8. As of October 2026

    Training a vision-language model

    As of October 2026, he is a Member of Technical Staff at Microsoft Research AI Frontiers, working with a team training a vision-language model and focusing on data quality and synthetic data generation.

Career

  1. NowMicrosoftMember of Technical Staff
  2. 2021–2025Institute for Advanced StudyPostdoctoral researcher
  3. 2020–2024Carnegie Mellon UniversityPostdoctoral researcher
  4. 2005–2014University of Sydney
  5. Lehigh UniversityPostdoctoral researcher
  6. Mathematical Sciences Research InstitutePostdoctoral researcher

Education

  1. 2017Pennsylvania State UniversityPhD

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Efficient Hyperelliptic Arithmetic Using Balanced Representation for Divisors2008 · Lecture notes in computer science · 44 citations
  2. A Lie algebra method for rational parametrization of Severi–Brauer surfaces2005 · Journal of Algebra · 34 citations
  3. An extension of Kedlaya’s algorithm for hyperelliptic curves2011 · Journal of Symbolic Computation · 23 citations
  4. Arithmetic and geometry of a K3 surface emerging from virtual corrections to Drell–Yan scattering2020 · Communications in Number Theory and Physics · 19 citations
  5. Explicit solution by radicals, gonal maps and plane models of algebraic curves of genus 5 or 62012 · Journal of Symbolic Computation · 15 citations

From OpenAlex, as of 21 October 2026.

  1. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Microsoft. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  2. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. 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.
  3. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Microsoft. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  4. Google has the highest combined h-index among AI organizationsIncludes Microsoft. 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.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

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

  1. About Me sites.google.com. Checked 21 October 2026.
  2. Introducing Totally Nonparallel Immersions arxiv.org. Checked 21 October 2026.
  3. Michael Harrison at Microsoft Research microsoft.com. Checked 21 October 2026.
  4. Phi-4 Technical Report - Microsoft Research microsoft.com. Checked 21 October 2026.
  5. Phi-4 Technical Report arxiv.org. Checked 21 October 2026.
  6. Fara-7B: An Efficient Agentic Model for Computer Use - Microsoft Research microsoft.com. Checked 21 October 2026.
  7. Phi-4-reasoning-vision and the lessons of training a multimodal reasoning model - Microsoft Research microsoft.com. Checked 21 October 2026.
  8. Phi-4-reasoning-vision-15B Technical Report arxiv.org. Checked 21 October 2026.
  9. Rho: A Foundation for Efficiently Adaptable VLA Models arxiv.org. Checked 21 October 2026.