PeopleofSuperintelligence

Researcher

Jian Jiao

Jian Jiao is a Partner Applied Science Manager on the Advertising team at Microsoft AI whose research spans web-scale deep learning, pretraining, sentence representations, and language-model monetization. At Microsoft Corporation in 2016, he coauthored the Deep Crossing modeling paper, followed by TwinBERT and BANG in 2020, HittER for knowledge graph embeddings in 2021, PromptBERT in 2022, and the AnnoLLM annotation system in March 2023. His coauthored 2024 paper on the Rho-1 model and Selective Language Modeling received a NeurIPS 2024 Best Paper Runner-up award. In 2026, he coauthored work on greedy information projection for LLM data selection, the MonBench monetization benchmark, and bias in LLMs for advertising relevance.

Now at
Microsoft

Timeline

  1. 2016

    Deep Crossing at KDD

    At Microsoft Corporation in 2016, he coauthors "Deep Crossing: Web-Scale Modeling without Manually Crafted Combinatorial Features" at KDD 2016 with Ying Shan, Thomas Hoens, Haijing Wang, Dong Yu, and JC Mao.

  2. 2020

    TwinBERT and BANG pretraining

    In February 2020, with Wenhao Lu and Ruofei Zhang, he coauthors TwinBERT to distill knowledge into twin-structured BERT models for efficient retrieval. In December 2020, he coauthors BANG with Weizhen Qi, Yeyun Gong, Yu Yan, and Weizhu Chen, bridging autoregressive and non-autoregressive generation with large-scale pretraining.

  3. 2021

    HittER for knowledge graph embeddings

    With Sanxing Chen, Xiaodong Liu, Jianfeng Gao, Ruofei Zhang, and Yangfeng Ji, he coauthors "HittER: Hierarchical Transformers for Knowledge Graph Embeddings" in 2021.

  4. 2022

    PromptBERT sentence embeddings

    In 2022, he coauthors PromptBERT, a contrastive learning method that uses prompts to improve BERT sentence embeddings.

  5. March 2023

    AnnoLLM for data annotation

    In March 2023, he coauthors AnnoLLM with Xingwei He, Zhenghao Lin, Yeyun Gong, A-Long Jin, Hang Zhang, and Chen Lin, introducing an LLM-powered annotation system that uses a two-step explain-then-annotate approach.

  6. 2024

    NeurIPS Best Paper Runner-up

    In 2024, he coauthors "Not All Tokens Are What You Need for Pretraining," which introduces Rho-1 and Selective Language Modeling to train selectively on useful tokens aligned with a desired distribution. NeurIPS names the paper a 2024 Best Paper Runner-up.

  7. 2026

    LLM data selection and monetization benchmarks

    In March 2026, he coauthors "Greedy Information Projection for LLM Data Selection" with Victor Ye Dong, Kuan-Yun Lee, Jiamei Shuai, Shen Liu, and Yi Liu. In 2026, he coauthors MonBench, a multi-task benchmark for evaluating LLMs in monetization contexts, and in October 2026 he coauthors a study of bias in LLMs for advertising relevance.

  8. As of October 2026

    Microsoft AI applied science manager

    In April 2026, he speaks at Stanford's Applications of Computational Math in Industry seminar on "The Next Monetization Stack: From Search and Ads to Commerce and Agents." He serves as a Partner Applied Science Manager on the Advertising team at Microsoft AI.

  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 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 Jian Jiao yet. Send us a free-licensed photo.

  1. Deep Learning kdd.org. Checked 22 October 2026.
  2. TwinBERT: Distilling Knowledge to Twin-Structured BERT Models for Efficient Retrieval arxiv.org. Checked 22 October 2026.
  3. BANG: Bridging Autoregressive and Non-autoregressive Generation with Large Scale Pretraining arxiv.org. Checked 22 October 2026.
  4. HittER: Hierarchical Transformers for Knowledge Graph Embeddings aclanthology.org. Checked 22 October 2026.
  5. PromptBERT: Improving BERT Sentence Embeddings with Prompts aclanthology.org. Checked 22 October 2026.
  6. AnnoLLM: Making Large Language Models to Be Better Crowdsourced Annotators arxiv.org. Checked 22 October 2026.
  7. Not All Tokens Are What You Need for Pretraining papers.nips.cc. Checked 22 October 2026.
  8. NeurIPS 2024 Awards neurips.cc. Checked 22 October 2026.
  9. Greedy Information Projection for LLM Data Selection microsoft.com. Checked 22 October 2026.
  10. Is Your Language Model Ready for Monetization Decisions? aclanthology.org. Checked 22 October 2026.
  11. A Systematic Investigation of Bias in Large Language Models for Advertising Relevance arxiv.org. Checked 22 October 2026.
  12. Applications of Computational Math in Industry, Spring 2026 icme.stanford.edu. Checked 22 October 2026.