PeopleofSuperintelligence

Researcher

Hao Cheng

Hao Cheng is a Principal Researcher at Microsoft Research and Affiliate Faculty at the University of Washington whose work spans biomedical language models, open-domain question answering, tool-using agents and human-LLM collaboration. After earning an MSc at the University of Alberta, he completed a PhD at the University of Washington, where he joined the Sounding Board team that won Amazon's inaugural Alexa Prize in 2017 and taught a graduate course on conversational AI in Spring 2019. His 2020 biomedical pretraining paper with Yu Gu, Robert Tinn and Hoifung Poon introduced the BLURB benchmark, and in 2021 it received a Best Paper Award in ACM Transactions on Computing for Healthcare while he published UnitedQA as first author. He coauthored Chameleon, Chain-of-Skills and LLaVA-Plus in 2023, along with the GRIN mixture-of-experts training method and Generative Adapter in 2024. In 2025, his paper CollabLLM on training language models for multiturn collaboration received an Outstanding Paper Award at ICML 2025.

Timeline

  1. 2017

    Winning the Alexa Prize

    He earns an MSc at the University of Alberta under Dale Schuurmans and Csaba Szepesvári and a PhD at the University of Washington with Mari Ostendorf. In 2017, as a University of Washington electrical engineering doctoral student working with team leader Hao Fang and Allen School doctoral students Elizabeth Clark, Ari Holtzman and Maarten Sap, he is on the Sounding Board team that wins Amazon's inaugural Alexa Prize and its $500,000 first prize.

  2. 2019

    Teaching conversational AI at Washington

    At the University of Washington in Spring 2019, he serves as instructor for the graduate course E596/LING 580 on conversational AI.

  3. 2020

    Biomedical language model pretraining

    In 2020, he coauthors a paper with Yu Gu, Robert Tinn and Hoifung Poon showing that pretraining language models from scratch on biomedical text gives substantial gains over continual pretraining of general-domain models. The work releases pretrained models and introduces the BLURB benchmark with a public leaderboard.

  4. 2021

    UnitedQA and a healthcare paper award

    He is first author of UnitedQA in 2021, written with Yelong Shen, Xiaodong Liu, Pengcheng He, Weizhu Chen and Jianfeng Gao, which combines an extractive and a generative reader to beat earlier models by 3.3 exact-match points on NaturalQuestions and 2.7 on TriviaQA. In 2021, his biomedical pretraining paper with Yu Gu, Robert Tinn, Hoifung Poon and colleagues receives a Best Paper Award in ACM Transactions on Computing for Healthcare.

  5. 2023

    Chameleon, Chain-of-Skills and LLaVA-Plus

    In April 2023, he coauthors Chameleon, a NeurIPS 2023 LLM system that composes tools such as vision models, web search engines and Python functions and reaches 86.54% accuracy on ScienceQA with GPT-4. In 2023, he also coauthors the configurable open-domain question-answering retriever Chain-of-Skills and, in November 2023, the multimodal assistant LLaVA-Plus, which keeps a skill repository of vision and vision-language models to activate tools for user requests.

  6. 2024

    GRIN mixture-of-experts training

    With 16 coauthors in September 2024, he introduces GRIN (GRadient-INformed MoE), a training method that uses sparse gradient estimation for expert routing so a model with 6.6B activated parameters matches a 14B dense model trained on the same data. In November 2024, he coauthors Generative Adapter with Tong Chen, Hao Fang, Patrick Xia, Xiaodong Liu, Benjamin Van Durme, Luke Zettlemoyer and Jianfeng Gao to contextualize language models in parameters with a single forward pass.

  7. 2025

    CollabLLM and an ICML award

    CollabLLM, which he coauthors in 2025 with Shirley Wu, Michel Galley, Baolin Peng, James Zou, Jure Leskovec, Jianfeng Gao and others, trains language models for multiturn collaboration using Multiturn-aware Rewards estimated from simulated conversations. In July 2025, the paper receives an Outstanding Paper Award at ICML 2025.

  8. As of October 2026

    Principal researcher at Microsoft

    He is a Principal Researcher at Microsoft Research and Affiliate Faculty at the University of Washington.

Career

  1. NowMicrosoft ResearchPrincipal Researcher
  2. NowUniversity of WashingtonAffiliate Faculty

Education

  1. University of AlbertaMSc
  2. University of WashingtonPhD

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Hao Cheng. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. 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.
  3. 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.
  4. 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.

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 Hao Cheng yet. Send us a free-licensed photo.

  1. Hao Cheng's Homepage sites.google.com. Checked 22 October 2026.
  2. UW students win Amazon's inaugural Alexa Prize for most engaging socialbot | UW News washington.edu. Checked 22 October 2026.
  3. UW Team wins international Amazon Alexa Prize for the design of conversational AI ece.uw.edu. Checked 22 October 2026.
  4. Domain-Specific Language Model Pretraining for Biomedical Natural Language Processing arxiv.org. Checked 22 October 2026.
  5. UnitedQA: A Hybrid Approach for Open Domain Question Answering arxiv.org. Checked 22 October 2026.
  6. Chameleon: Plug-and-Play Compositional Reasoning with Large Language Models arxiv.org. Checked 22 October 2026.
  7. Chain-of-Skills: A Configurable Model for Open-domain Question Answering arxiv.org. Checked 22 October 2026.
  8. LLaVA-Plus: Learning to Use Tools for Creating Multimodal Agents arxiv.org. Checked 22 October 2026.
  9. GRIN: GRadient-INformed MoE arxiv.org. Checked 22 October 2026.
  10. Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass arxiv.org. Checked 22 October 2026.
  11. CollabLLM: From Passive Responders to Active Collaborators arxiv.org. Checked 22 October 2026.
  12. Congratulations to the #ICML2025 award winners! - AIhub aihub.org. Checked 22 October 2026.
  13. Hao Cheng at Microsoft Research microsoft.com. Checked 22 October 2026.