📝 Selected Publications
🎓 First-Author Top-Tier Conference Papers During My Master's
ICML 2026Can LLMs Reason Over Multimodal Structures Without Wasting Tokens?
Designed an entropy-guided smart adapter that focuses on the instruction-related parts of multimodal inputs, solving the longstanding gap of LLMs treating all positions as equally important. Achieves top-1 on 17 out of 18 benchmarks.
ICLR 2026How Can LLMs Understand Dynamic Modalities Without Fixed-Size Bottlenecks?
A dynamic connector that scales representation slots based on complexity, bridging the gap of fixed-size connectors that lose structural detail for large entities. The model ranks #1 on 20 of 21 understanding tasks and trains 3.5× faster under equivalent settings.
NeurIPS 2025Can Multimodal Foundation Models Learn Reliable Representations from Noisy 3D?
Built a multimodal encoder that first stabilizes noisy 3D shape data into a reliable prior, solving the longstanding instability in multimodal embedding that incorporates 3D geometry. Ranks #1 on 22/29 benchmarks with up to 27% better accuracy.
🎓 Additional Work and Co-Authored Publications
- Piccolo2: General Text Embedding with Multi-Task Hybrid Loss Training. Junqin Huang, Zhongjie Hu, Zihao Jing, Mengya Gao, Yichao Wu. SenseTime Technical Report, 2024.
- Attention with Routed-Memory for Learnable Sparse Control. Qiuhao Zeng, ..., Zihao Jing, et al.. ICML 2026.
- SAGA: Structural Aggregation Guided Alignment for Multiview Graph Domain Adaptation. Ruiyi Fang, Jingyu Zhao, ..., Zihao Jing, et al.. ICLR 2026.
📄 Research Funding
Digital Research Alliance of Canada — RRG Competition Grant 2026
Received 25 RGU-years (≡ 5×A100-80GB GPU-years) on Canada's national supercomputing cluster via the Resources for Research Groups Competition. Estimated commercial value: ~US$80K.
💻 Internships
- Developed Piccolo-GPT, a dual-function LLM supporting both text embedding and generation within a single architecture.
- Trained general-purpose text embedding models (Piccolo2); top-1 on C-MTEB (May 2024).
- Fine-tuned a ~100B LLM for vertical livestream marketing for Sina Weibo.
- Implemented LLM-based denoising and sentiment analysis pipeline for Budweiser; reduced cost >13%.
- Led evaluation and deplopying of super-resolution models for commercialization.
- Contributed to RunGPT of LlamaIndex with our super-resolution pipeline.
📖 Education
- Sep 2024 – Apr 2026MSc, Computer Science, Western University. London, Canada. Supervised by Prof. Pingzhao Hu.Graduated with Honor
- Sep 2019 – Jun 2024BEng, Software Engineering, Beihang University. Beijing, China. Graduated with Honor.
🛠 Technical Skills
Post-training of large multimodal models (SFT, RL). First-author papers at NeurIPS, ICLR, ICML on structured reasoning and multimodal alignment.
HPC and cloud-scale training infrastructure. Trained models up to ~100B parameters on national supercomputing clusters.
Strong in data structures, algorithms, and full-stack development. Led 3 large production websites deployed in Shanghai G60, Beihang, and Chaoyang.
9+ models and 8+ datasets published with 2k+ total downloads. Experience with multi-agent systems, vector databases, and RAG pipelines.
🎤 Selected Talks / Presentations
🎖 Honors & Awards
- 2023Silver Prize — Feng Ru Cup Science and Innovation Competition(University)
- 2021Third Prize — 13th National Undergraduate Mathematics Competition(National)
- 2021Third Prize — 32nd Beijing Undergraduate Mathematics Competition(Province)
- 2021Outstanding Student Leader Award(University)
- 2021Third Prize — Physics Academic Competition(University)
- 2020H Prize — Mathematical Contest in Modeling (MCM)(International)
👁 Academic Service
- ICML 2026 — Conference Reviewer, 6 papers reviewed🏅 Gold Reviewer Award
- ICLR 2026 — Conference Reviewer, 5 papers reviewed
- NeurIPS 2026 — Conference Reviewer
- TSALM @ ICLR 2026; AI4GOOD, FMSD, GFM @ ICML 2026 — Workshop Reviewer
- ACM TKDD — Journal Reviewer
- IEEE TNNLS — Journal Reviewer
- NeurIPS 2025 — Conference Volunteer, San Diego
🧑🏫 Mentorship
Teaching Assistant at Western University
- COMPSCI 2211 — Software Tools and Systems Programming (C and Linux) — Fall 2024 · Fall 2025
- COMPSCI 3305 — Operating Systems — Winter 2025 · Winter 2026
👨🏫 Research Values from My Mentors
“Write a paper such that reviewers neither can nor need to ask questions.”
“Read 20+ related papers before discussing ideas with me.”
“A paper should present clear, well-aligned challenges and novelties.”
“If you dig one direction deeper, you will always uncover novelty.”