About Me

余宏伟Hongwei Yu I am currently a Ph.D. candidate in Computer Science and Technology at University of Science and Technology Beijing (USTB) (Sep. 2022–present). My research focuses on Post-Training, MLLM, and AIGC.

During my Ph.D., I interned at DreamX Team, Alibaba Amap, StepFun, and JD Retail. I am currently with JD Retail's TGT Program, focusing on video generation and seeking full-time opportunities starting in Fall 2027; please feel free to contact me.

Email Google Scholar GitHub

Education

Sep. 2022 – PresentPh.D. candidate, University of Science and Technology Beijing, Computer Science and Technology, Beijing.
Recipient of the National Scholarship for Doctoral Students.
Ph.D. advisor: Prof. Jiansheng Chen.
Research group: 3D Image Lab, led by Prof. Huimin Ma.
Sep. 2018 – Jun. 2022B.Eng., University of Science and Technology Beijing, Computer Science and Technology, Beijing.
Recipient of the National Scholarship for Undergraduate Students.

News

Oct. 2026My first-authored paper Hybrid Reinforcement Learning for One-step Degraded Infrared and Visible Image Fusion was accepted as a Spotlight at NeurIPS 2026 (292 / 30,709; 0.9%).
Aug. 2026Our co-first-authored paper Skill-Guided Continuation Distillation for GUI Agents was accepted to EMNLP 2026.
Aug. 2026Our co-first-authored paper Decoupling Cross-Modality Manifold Discrepancy: Leveraging Visible Diffusion Priors for Infrared Super-Resolution was accepted to ACM MM 2026.
May 2026Joined JD Retail's TGT Program to work on multimodal image and video generation foundation models.
Jan. 2026Joined the Foundation Model Group at StepFun to work on GUI-agent post-training.
2026My first-authored paper MEDUSA and co-first-authored paper FactGuard were accepted to ICML 2026.
Sep. 2025Joined the DreamX Team, Alibaba Amap to work on AIGC research.
2025My first-authored paper DADet was accepted as a Spotlight at ICCV 2025.

Work / Internships

May 2026 – PresentJD Retail, TGT Program, Beijing. Multimodal image and video generation foundation models.
Jan. 2026 – May 2026StepFun, Foundation Model Group, Beijing. GUI-agent post-training.
Sep. 2025 – Jan. 2026DreamX Team, Alibaba Amap, Beijing. AIGC algorithms.

Selected / First-author Publications

Bold indicates Hongwei Yu. * indicates equal contribution.

Hybrid Reinforcement Learning for One-step Degraded Infrared and Visible Image Fusion

NeurIPS 2026 Spotlight · First author · 0.9% (292 / 30,709)

Hybrid Reinforcement Learning for One-step Degraded Infrared and Visible Image Fusion

H. Yu, D. Zha, Y. Bai, J. Li, L. Sun, X. Chu.

Skill-Guided Continuation Distillation for GUI Agents

EMNLP 2026 Co-first author

Skill-Guided Continuation Distillation for GUI Agents

Z. Fan*, H. Yu*, Y. Shen, H. Yan, G. Peng, T. Peng, Y. Zhang, X. Zhang, K. Tan, et al.

MEDUSA

ICML 2026 First author

MEDUSA: Motion Elimination in Diffusion Using Spectral Attack

H. Yu, et al.

FactGuard

ICML 2026 Co-first author

FactGuard: Agentic Video Misinformation Detection via Reinforcement Learning

Z. Li*, H. Yu*, H. Jiang, Q. Sheng, Y. Xu, B. Bi, Y. Li, Z. Yuan, Y. Cai, Z. Wang.

Decoupling Cross-Modality Manifold Discrepancy

ACM MM 2026 Co-first author

Decoupling Cross-Modality Manifold Discrepancy: Leveraging Visible Diffusion Priors for Infrared Super-Resolution

Y. Hua*, H. Yu*, J. Li, Q. Liu, H. Ma, J. Chen.

DADet

ICCV 2025 Spotlight · First author

DADet: Safeguarding Image Conditional Diffusion Models against Adversarial and Backdoor Attacks via Diffusion Anomaly Detection

H. Yu, X. Ding, J. Li, J. Wang, Y. Zhang, R. Wang, H. Ma, J. Chen.

A²RNet

AAAI 2025 Co-first author

A²RNet: Adversarial Attack Resilient Network for Robust Infrared and Visible Image Fusion

J. Li*, H. Yu*, J. Chen, X. Ding, J. Wang, J. Liu, B. Zou, H. Ma.

Kaleidoscopic Background Attack

ICCV 2025 Spotlight · Co-first author

Kaleidoscopic Background Attack: Disrupting Pose Estimation with Multi-Fold Radial Symmetry Textures

X. Ding*, H. Yu*, J. Li, F. Li, Y. Shang, B. Zou, H. Ma, J. Chen.

Step Vulnerability Guided Mean Fluctuation Adversarial Attack

AAAI 2024 First author

Step Vulnerability Guided Mean Fluctuation Adversarial Attack against Conditional Diffusion Models

H. Yu, J. Chen, X. Ding, Y. Zhang, T. Tang, H. Ma.

Defending against Universal Patch Attacks by Restricting Token Attention in Vision Transformers

ICASSP 2023 First author

Defending against Universal Patch Attacks by Restricting Token Attention in Vision Transformers

H. Yu, J. Chen, H. Ma, C. Yu, X. Ding.

Other Publications

IEEE TMM 2026 Multi-Scale Spatial Channel Joint Representation for General Multi-Modality Image Fusion With Self-Supervision, J. Li, J. Chen, J. Liu, H. Yu, X. Ding, H. Ma.

ICME 2024 Invisible Pedestrians: Synthesizing Adversarial Clothing Textures To Evade Industrial Camera-Based 3D Detection, X. Ding, H. Yu, J. Chen, J. Wang, J. Du, H. Ma.

ICASSP 2024 Enhancing Adversarial Transferability in Object Detection with Bidirectional Feature Distortion, X. Ding, J. Chen, H. Yu, Y. Shang, B. Zou, H. Ma.

AAAI 2024 Transferable Adversarial Attacks for Object Detection Using Object-Aware Significant Feature Distortion, X. Ding, J. Chen, H. Yu, Y. Shang, Y. Qin, H. Ma.

KBS 2024 Image Paragraph Captioning with Topic Clustering and Topic Shift Prediction, T. Tang, J. Chen, Y. Huang, H. Ma, Y. Zhang, H. Yu.

For a complete publication list, please visit Google Scholar.

Academic Service

Conference and Journal Reviewing

ARR (ACL / EMNLP / NAACL)
CVPR, ICCV, ECCV, NeurIPS, AAAI, ACM-MM, ICML

Honors and Awards

Gold Reviewer Award @ ICML 2026
National Scholarship for Doctoral Students
National Scholarship for Undergraduate Students
Huawei Future Star Scholarship
First Prize, National Computer Game Competition
Merit Graduate Student / Merit Student
Outstanding Graduate of USTB