Haonan Wang

Also known as 王淏楠

Haonan Wang is an AI researcher whose work covers neural document retrieval, long-context language-model training and reinforcement learning for multimodal generation. He studied computer science and statistics at the University of Illinois Urbana-Champaign from 2017 to 2020, then spent 2020 to 2021 as a PhD student there under Jingrui He. As a Microsoft Research Asia intern in 2022, he co-authored the Neural Corpus Indexer, which NeurIPS 2022 listed among its Outstanding Papers. At the National University of Singapore he completed a PhD under Kenji Kawaguchi in February 2026, with an ICML 2024 oral paper on copyright backdoors in diffusion models and, from an internship at Sea AI Lab, the AnchorAttention method for long-context training. Since March 2026, he is a Principal Research Scientist at Tencent Hunyuan, where he leads the open-source UniRL framework.

Now at: Tencent
Badges: Bridge

Timeline

2017–2020

Undergraduate at Illinois

He studies computer science and statistics at the University of Illinois Urbana-Champaign from 2017 to 2020. As an undergraduate, he does research with Jiawei Han.
2020–2021

PhD student under Jingrui He

He spends 2020 to 2021 as a computer science PhD student at the University of Illinois Urbana-Champaign, advised by Jingrui He. In 2020, he interns at Alibaba in Beijing on item retrieval and recommendation, and his first-author paper Controllable Gradient Item Retrieval appears at WWW 2021.
2022

The Neural Corpus Indexer

From February 2022 to May 2022, he is a research intern at Microsoft Research Asia in Beijing, working on neural retrieval and corpus indexing. He co-authors A Neural Corpus Indexer for Document Retrieval, which proposes a sequence-to-sequence network that generates relevant document identifiers directly for a query.
2022

A NeurIPS outstanding paper

NeurIPS 2022 lists A Neural Corpus Indexer for Document Retrieval among its Outstanding Papers, with him among the co-authors. The conference announces its award-winning papers in November 2022.
2023–2026

PhD with Kenji Kawaguchi

He begins a computer science PhD at the National University of Singapore's School of Computing in 2023, advised by Kenji Kawaguchi. He completes the degree in February 2026.
2024

Copyright backdoors in diffusion models

In 2024, he is first author of The Stronger the Diffusion Model, the Easier the Backdoor, presented at ICML 2024, which selects it as an oral. The paper proposes SilentBadDiffusion, a data-poisoning backdoor that induces copyright infringement in text-to-image diffusion models, and finds that more sophisticated models are easier to attack.
June 2024–May 2025

AnchorAttention at Sea AI Lab

From June 2024 to May 2025, he is a research intern at Sea AI Lab in Singapore and leads the AnchorAttention project on long-context training. His paper When Precision Meets Position: BFloat16 Breaks Down RoPE in Long-Context Training, posted in November 2024 and accepted to TMLR in February 2025, shows that BFloat16 makes RoPE deviate from its intended relative positional encoding at long context lengths, and proposes AnchorAttention as the fix.
Since March 2026

Leading UniRL at Tencent Hunyuan

In March 2026, he joins Tencent Hunyuan in Singapore as a Principal Research Scientist, leading reinforcement learning for agentic multimodal generation and the open-source UniRL framework. UniRL applies one RL post-training loop across multimodal model families.