Aug. 2022 - Jul. 2026
The Chinese University of Hong Kong
Ph.D. in Computer Science. I am part of the DVLab at CUHK, working with Prof. Jiaya Jia.
Jingyao Li（李靖瑶）
I am a Ph.D. candidate at The Chinese University of Hong Kong, advised by Prof. Jiaya Jia. I completed my undergraduate study at Xi'an Jiaotong University in the Special Class for the Gifted Young (少年班) and the Qian Xuesen Honor Class (钱学森班).
I am currently interested in foundation large language models, especially mid-training and agentic reinforcement learning.
Education
Sept. 2018 - Jul. 2022
Xi'an Jiaotong University
B.Eng. in Automation. Special Class for the Gifted Young (少年班) and Qian Xuesen Honor Class (钱学森班).
Experience
Apr. 2024 - Nov. 2024
Huawei Noah's Ark Lab · Research Intern
Worked on query-as-computation for large models and first-authored QuickLLaMA: Query-aware Inference Acceleration for Large Language Models.
Dec. 2024 - Jul. 2025
Alibaba Group's Tongyi Lab · Research Intern
Participated in large-scale model pre-training, led work on a continual pre-training QA construction pipeline, and contributed data and training solutions to Qwen3.
Publications
Rethinking Out-of-Distribution Detection: Masked Image Modeling is All You Need.
CVPR 2023BAL: Balancing Diversity and Novelty for Active Learning.
TPAMI 2023MOODv2: Masked Image Modeling for Out-of-Distribution Detection.
TPAMI 2024TagCLIP: Improving Discrimination Ability of Open-Vocabulary Semantic Segmentation.
TPAMI 2024QuickLLaMA: Query-aware Inference Acceleration for Large Language Models.
COLING 2025VLPose: Bridging the Domain Gap in Pose Estimation with Language-Vision Tuning.
TPAMI 2025Logits-Based Finetuning.
EMNLP 2025Other Publications
MR-BEN: A Comprehensive Meta-Reasoning Benchmark for Large Language Models.
NeurIPS 2024DAPE: Data-Adaptive Positional Encoding for Length Extrapolation.
NeurIPS 2024Visionzip: Longer is Better but Not Necessary in Vision Language Models.
CVPR 2025DreamOmni: Unified Image Generation and Editing.
CVPR 2025DAPE V2: Process Attention Score as Feature Map for Length Extrapolation.
ACL 2025Lyra: An Efficient and Speech-Centric Framework for Omni-Cognition.
ICCV 2025DLoFT: Gradient-Decoupled Fine-Tuning for Generalizable Long Chain-of-Thought Reasoning.
NeurIPS 2025Unlocking Token Rewards via Training-Free Reward Attribution.
CVPR 2026DreamOmni2: Multimodal Instruction-based Editing and Generation.
CVPR 2026Understanding the Mixture-of-Expert with Nadaraya-Watson Kernel.
ICLR 2026MoTCoder: Elevating Large Language Models with Modular of Thought for Challenging Programming Tasks.
ArxivRoboCoder: Robotic Learning from Basic Skills to General Tasks with Large Language Models.
ArxivEnd-to-End QA Construction Pipeline for Continual Pre-training of Large Language Models.
ArxivDynamicBench: Evaluating Real-Time Report Generation in Large Language Models.
ArxivMixture-of-Scores: Robust Image-Text Data Quality Score via Three Lines of Code.
ArxivMGM-Omni: Scaling Omni LLMs to Personalized Long-Horizon Speech.
ArxivAnswer-Consistent Chain-of-thought Reinforcement Learning For Multi-modal Large Langauge Models.
ArxivMulti-Scale Group Relative Policy Optimization for Large Language Models.
ArxivTeaching & Honors
Teaching
- (CENG2010) Digital Logic Design Laboratory, 2023 Spring
- (CSCI3250) Computers and Society, 2022 Fall
Awards
- National Scholarship, 2020
- Meritorious Winner, American Mathematical Contest in Modeling, 2019
- First Prize, China Undergraduate Physics Tournament, 2018
- Second Prize, Xi'an Jiaotong University Challenge Cup, 2019
Leadership
- Vice President, Youth Volunteers Association, 2019-2020
- Minister of the Secretariat, Youth Volunteers Association, 2018-2019
- Administrative Department Member, Youth Volunteers Association, 2017-2018
- Secretary of the Student Union, Suzhou Middle School, 2016-2017