About Me [CV]

I am a Ph.D. candidate in Computer Engineering at Northeastern University, where I work in the SMILE Lab with Prof. Yun Raymond Fu (Member of Academia Europaea, Fellow of ACM, AAAS, AAAI, IEEE, IAPR). I earned my B.Eng. and M.S. at Xidian University, where Prof. Xuefeng Liang advised my master's research. During two internships at Adobe Research, I studied how vision-language models understand the structure of graphic designs and how their critiques can guide an agent's layout edits.

Research interests Multimodal LLMs & Agents | Long-Video Understanding | Hallucination Detection & Mitigation | Agent Reasoning, Reflection, Tool Use & Self-Improvement

Actively seeking internship opportunities for Spring/Summer 2027. Feel free to reach out for collaborations or any inquiries!


News


Experience

SMILE Lab, Northeastern University, Boston
Ph.D. candidate in Computer Engineering, Sep. 2024 – Present (expected graduation May 2029)
Northeastern University
Adobe Research, San Jose
Research Intern, May 2026 – Nov. 2026
Adobe Research
Adobe Research, San Jose
Research Intern, May 2025 – Nov. 2025
Adobe Research
Kyoto University, Kyoto
Visiting Student, Sep. 2023 – Mar. 2024
Kyoto University
Xidian University, Xi'an
M.S. in Computer Science, Sep. 2021 – Jun. 2024
B.Eng. in Intelligence Science and Technology, Sep. 2017 – Jun. 2021
Xidian University

Publications [Google Scholar]

Under review
GRACE teaser comparing human-designed posters with alternative layouts and showing how human and model preferences differ.
When Plausibility Fails: Learning Human-Aligned Layout Evaluation with VLMs
Yiyang Huang, Zhaowen Wang, Shraman Pramanick, Yun Fu
TL;DR: GRACE critiques layout designs and scores an editing agent’s revisions. GRACE-27B reached 83.7% accuracy on DesignJudge, a dataset of 7,475 layout preference pairs.
Preprint
RACER framework: query comprehension, tool-based retrieval, and reflection
RACER: Reflective Agent Coupling Query Interpretation and Tool-Based Retrieval for Frame Selection in Long Video Understanding
Yiyang Huang, Yitian Zhang, Yizhou Wang, Jianglin Lu, Qihua Dong, Hailing Wang, Huimin Zeng, Mingyuan Zhang, Yun Fu
TL;DR: RACER breaks a question into retrieval queries, selects video frames, and refines its queries using visual feedback. It improved accuracy by 4.3–6.6 percentage points over uniform sampling across three long-video benchmarks.
ECCV 2026
Thumbnail: Beyond Atomic Layouts paper
Beyond Atomic Layouts: Compositional Design Understanding with Vision-Language Models
Yiyang Huang, Zhaowen Wang, Simon Jenni, Jing Shi, Yitian Zhang, Yizhou Wang, Yun Fu
TL;DR: MASON uses design metadata and 3D structural cues to help vision-language models interpret layered layouts. CoDeLayout benchmarks this task with about 20,000 examples.
ACL 2026
Thumbnail: Video LLM Hallucination Survey paper
Distorted or Fabricated? A Survey on Hallucination in Video LLMs
Yiyang Huang, Yitian Zhang, Yizhou Wang, Mingyuan Zhang, Liang Shi, Huimin Zeng, Yun Fu
TL;DR: This survey organizes hallucinations in video LLMs by type and reviews the benchmarks and methods used to detect and reduce them.
IEEE FG 2026
Thumbnail: IDEA facial expression generation paper
Capturing Individual Differences of Facial Expression for Authentic Expression Generation
Liang Shi, Yiyang Huang, Yun Fu
TL;DR: Trained a CLIP-style contrastive model with identity and expression modalities to guide text-to-image generation toward more authentic, individual-specific expressions.
ICLR 2026
Thumbnail: SHIELD hallucination mitigation paper
SHIELD: Suppressing Hallucinations in LVLM Encoders via Bias and Vulnerability Defense
Yiyang Huang, Liang Shi, Yitian Zhang, Yi Xu, Yun Fu
TL;DR: SHIELD combines inference-time token edits with adversarial contrastive decoding. It reduced captioning hallucinations by 27% relative to baseline across multiple vision-language model families.
EMNLP 2025
Thumbnail: D-CoDe video understanding paper
D-CoDe: Scaling Image-Pretrained VLMs to Video via Dynamic Compression and Question Decomposition
Yiyang Huang, Yizhou Wang, Yun Fu
TL;DR: D-CoDe breaks questions into parts and compresses video frames and tokens. With LLaVA-NeXT-7B, it improved question-answering accuracy on first-person videos by 29% relative to baseline.
ICASSP 2025
Thumbnail: LipReading low-resource languages paper
LipReading for Low-resource Languages by Language Dynamic LoRA
Shuai Zou, Xuefeng Liang, Yiyang Huang
TL;DR: Developed dynamic LoRA for meta lip shapes and multilingual instruction tuning to improve cross-lingual lipreading in low-resource settings.
ACM MM 2021
Thumbnail: CALLip lipreading paper
CALLip: Lipreading using Contrastive and Attribute Learning
Yiyang Huang, Xuefeng Liang, Chaowei Fang
TL;DR: Proposed CALLip, leveraging attribute learning to normalize cross-speaker variation and audio-visual contrastive learning to mitigate viseme confusion.
Under Review
Thumbnail: StreamVLM paper
StreamVLM: Adapting Offline Vision-Language Models for Streaming Understanding
Hailing Wang, Jianglin Lu, Xu Ma, Yiyang Huang, Yun Fu
TL;DR: Adapts offline VLMs to streaming video understanding with event-based KV-cache memory construction and KNN-based retrieval.
Preprint
Thumbnail: Rethinking Fine-Tuning for VLMs paper
Rethinking Fine-Tuning: Unlocking Hidden Capabilities in Vision-Language Models
Mingyuan Zhang, Yue Bai, Yifan Wang, Yiyang Huang, Yun Fu
TL;DR: Applied MFT to VLMs: learnable gating reorganizes subnetworks without weight updates; outperforms LoRA and full fine-tuning.
Preprint
MASKerade framework compared with conventional mixture-of-experts
MASKerade: Token-Routed Mask Experts for Dense-to-MoE Upcycling
Mingyuan Zhang, Yue Bai, Zhongruo Wang, Yupin Huang, Yiyang Huang, Hailing Wang, Huimin Zeng, Yun Fu
TL;DR: Converts dense models into mixture-of-experts by jointly learning sparse expert masks and token-level routing over frozen feed-forward weights.

Academic Service

Conference Reviewer ICLR, NeurIPS, ACL, EMNLP, WWW, IEEE FG
Journal Reviewer ACM TKDD

Honors & Awards

2022 Outstanding Student, Xidian University
2021 National Scholarship, China
2021 Undergraduate Computer Design Competition (1st Prize, National Level), China
2019 RoboMaster National Robotics Competition (2nd Prize), China
2019 ICRA AI Challenge (3rd Prize)

Teaching Experience

Fall 2025 TA — DS 5110 Essentials of Data Science
Spr. & Fall 2026 TA — DS 5020 Fundamentals of Linear Algebra and Probability

Contact