PeopleofSuperintelligence

Researcher

Harkirat Behl

Harkirat Behl is a machine-learning researcher whose work spans adaptive learning, synthetic data, computer vision and language models. He earned a BTech at IIT Kanpur in 2018 and a PhD at Oxford in 2021, then coauthored early work on few-shot learning and simulator-based training data. His later work includes Phi-4 and Phi-4-reasoning, 14-billion-parameter models for language and complex reasoning tasks. In March 2026, he coauthored ML-AutoResearch; as of October 2026, he is a Principal Researcher at Microsoft Research.

Timeline

  1. 2018–2021

    IIT Kanpur and Oxford

    He earns a BTech at the Indian Institute of Technology, Kanpur, in 2018 and completes a PhD at the University of Oxford in 2021, advised by Philip Torr and M. Pawan Kumar.

  2. 2019

    Alpha MAML

    In 2019, he coauthors Alpha MAML with Atılım Güneş Baydin and Philip Torr at the ICML AutoML Workshop. The method adds online adaptation of learning and meta-learning rates to reduce hyperparameter tuning.

  3. 2020

    AutoSimulate for synthetic data

    He coauthors AutoSimulate in 2020, which optimizes a simulator through a differentiable approximation of the data-generation objective. The team reports up to 30× less training-data generation and 8.7% higher accuracy on real-world test data.

  4. 2022

    Neural-Sim and synthetic data

    In 2022, he coauthors Neural-Sim, a differentiable synthetic-data pipeline that uses Neural Radiance Fields in a closed loop with a target application's loss function.

  5. 2024

    Phi-4 language model

    In 2024, he coauthors Phi-4, a 14-billion-parameter language model whose training recipe centers on data quality and incorporates synthetic data throughout training.

  6. 2025

    Phi-4-reasoning models

    In April 2025, he coauthors Phi-4-reasoning, a 14-billion-parameter model for complex reasoning tasks. The report also introduces Phi-4-reasoning-plus, which uses outcome-based reinforcement learning to generate longer reasoning traces.

  7. As of October 2026

    Principal Researcher at Microsoft Research

    In March 2026, he coauthors ML-AutoResearch, a pipeline that creates end-to-end machine-learning tasks and uses trajectories from those tasks to train research agents. As of October 2026, he is a Principal Researcher at Microsoft Research.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Sources and credits

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

  1. Harkirat Behl harkiratbehl.github.io. Checked 19 October 2026.
  2. Neural-Sim: Learning to Generate Training Data with NeRF arxiv.org. Checked 19 October 2026.
  3. Phi-4 Technical Report arxiv.org. Checked 19 October 2026.
  4. Harkirat Behl at Microsoft Research microsoft.com. Checked 19 October 2026.
  5. Phi-4-reasoning Technical Report arxiv.org. Checked 19 October 2026.
  6. ML-AutoResearch: Training Machine Learning Research Agents with Automatically Generated Environments arxiv.org. Checked 19 October 2026.