PeopleofSuperintelligence

Researcher

Jiantao Jiao

Jiantao Jiao is a machine learning researcher whose work includes reinforcement learning and foundation models. He earned engineering degrees at Tsinghua University and Stanford University, completed his Stanford PhD in 2018, and joined UC Berkeley in 2018. His coauthored research introduced the TRADES defense against adversarial examples in 2019 and the Starling-7B language model in 2023. He co-founded Nexusflow, served as its CEO, and became a director of research and distinguished scientist at NVIDIA after it acquired the company. As of October 2026, he leads NVIDIA's Nemotron data, scaling and evaluation team, holds adjunct and affiliated assistant professorships at Berkeley, and co-directs Berkeley's CLIMB center.

Works
144
Citations
3,131
h-index
22
Badges

Publication numbers from OpenAlex, as of 28 October 2026.

Timeline

  1. 2012–2014

    Engineering degrees at Tsinghua and Stanford

    He earns a B.Eng. in electronic engineering at Tsinghua University in 2012 and an M.Eng. in electrical engineering at Stanford University in 2014.

  2. 2016–2017

    Teaching statistical signal processing

    He shares the ISITA Student Paper Award in 2016. With his advisor, Tsachy Weissman, he co-designs and co-teaches Stanford's graduate course EE378A (Statistical Signal Processing) in 2016 and 2017.

  3. 2018

    Stanford PhD and Berkeley faculty

    He completes his PhD at Stanford University in 2018. In 2018, he joins UC Berkeley's Department of Electrical Engineering and Computer Sciences as an assistant professor, studying causal relationships with applications in health and social sciences.

  4. 2019

    The TRADES defense

    In 2019, he coauthors “Theoretically Principled Trade-off between Robustness and Accuracy,” published at ICML. The paper introduces TRADES, a method for defending machine learning models against adversarial examples.

  5. 2021

    Offline reinforcement learning and imitation

    He coauthors “Bridging Offline Reinforcement Learning and Imitation Learning: A Tale of Pessimism” at NeurIPS in 2021. The work develops a framework for learning from fixed datasets with different levels of expert behavior and analyzes a lower confidence bound algorithm.

  6. September 2023

    Co-founding Nexusflow

    By September 2023, he is a co-founder and CEO of Nexusflow, a generative AI company for cybersecurity. The company raises $10.6 million in seed financing led by Point72 Ventures in September 2023, developing AI models that enterprises can use without leaking sensitive data.

  7. November 2023

    The Starling-7B language model

    In November 2023, he helps introduce Starling-7B, an open language model trained with reinforcement learning from AI feedback. The model uses Nectar, a dataset of responses ranked by GPT-4, and a reward training and policy tuning pipeline.

  8. By 2025

    Nexusflow acquisition by NVIDIA

    By November 2025, NVIDIA acquires Nexusflow, where he serves as CEO and leads work on foundation models and agentic systems. He is a director of research and distinguished scientist at NVIDIA by November 2025.

  9. As of October 2026

    Berkeley faculty and NVIDIA research

    As of October 2026, he is an adjunct assistant professor in UC Berkeley's Department of Electrical Engineering and Computer Sciences and an affiliated assistant professor in its Department of Statistics. He co-directs the Center for the Theoretical Foundations of Learning, Inference, Information, Intelligence, Mathematics and Microeconomics at Berkeley (CLIMB). At NVIDIA, he is a director of research and distinguished scientist who leads the Nemotron data, scaling and evaluation team, which develops methods to predict model capabilities from early training signals and guide data and training decisions.

Career

  1. NowNVIDIADirector of Research and Distinguished Scientist
  2. NowUniversity of California, BerkeleyAdjunct Assistant Professor
  3. NowUniversity of California, BerkeleyAffiliated Assistant Professor
  4. NowUniversity of California, BerkeleyCo-director of the Center for the Theoretical Foundations of Learning, Inference, Information, Intelligence, Mathematics and Microeconomics
  5. NowNVIDIALeader of the Nemotron data, scaling and evaluation team
  6. From 2018University of California, BerkeleyAssistant Professor
  7. NexusflowCo-founder
  8. NexusflowCEO

Education

  1. 2018Stanford UniversityPhD in electrical engineering
  2. 2014Stanford UniversityM.Eng. in electrical engineering
  3. 2012Tsinghua UniversityB.Eng. in electronic engineering

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

Checked career sources: people.eecs.berkeley.edu, eecs.berkeley.edu, isl.stanford.edu.

Colleagues

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

Sources and credits

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

  1. Jiantao Jiao (Stanford) bliss.eecs.berkeley.edu. Checked 28 October 2026.
  2. Jiantao Jiao people.eecs.berkeley.edu. Checked 28 October 2026.
  3. Six new EECS faculty welcomed in 2018 eecs.berkeley.edu. Checked 28 October 2026.
  4. Theoretically Principled Trade-off between Robustness and Accuracy arxiv.org. Checked 28 October 2026.
  5. Bridging Offline Reinforcement Learning and Imitation Learning: A Tale of Pessimism arxiv.org. Checked 28 October 2026.
  6. Nexusflow Announces $10.6M Seed to Revolutionize Cybersecurity gunder.com. Checked 28 October 2026.
  7. Starling-7B: Increasing LLM Helpfulness & Harmlessness with RLAIF starling.cs.berkeley.edu. Checked 28 October 2026.
  8. Jiantao Jiao | ISL Colloquium isl.stanford.edu. Checked 28 October 2026.
  9. Jiantao Jiao | EECS at UC Berkeley www2.eecs.berkeley.edu. Checked 28 October 2026.
  10. Jiantao Jiao | Department of Statistics statistics.berkeley.edu. Checked 28 October 2026.
  11. People | CLIMB climb.berkeley.edu. Checked 28 October 2026.
  12. Author: Jiantao Jiao | NVIDIA Technical Blog developer.nvidia.com. Checked 28 October 2026.