Boxue Yang (杨博学)

Boxue Yang

Shanghai Jiao Tong University

About Me

I am an undergraduate student majoring in Software Engineering at the School of Computer Science, Shanghai Jiao Tong University (SJTU) (Class of 2022). My current GPA is 92.5/100, ranking 5/88 in my major. Due to outstanding academic performance and research potential, I have been awarded the National Scholarship (Top 1%) and have been admitted to the School of Computer Science at SJTU to pursue a Master's degree (starting Fall 2026).

My research interests primarily focus on Multimodal Large Language Models and Video Understanding. I am passionate about exploring frontier AI technologies, specifically focusing on empowering models with stronger scientific reasoning and real-time interaction capabilities. I have led and participated in several challenging research projects, including scientific multimodal models and streaming video models.

Feel free to contact me via email to discuss academic ideas or explore collaboration opportunities.

🔥 News

  • 2026.02: 🎉 Our papers EvoStreaming and Robust Graph Representation have been submitted to ICML 2026.
  • 2026.01: 🚀 Released Innovator-VL, a multimodal large language model designed for scientific discovery.
  • 2025.12: 🏆 Awarded the National Scholarship for the 2024-2025 academic year (Top 1%).
  • 2024.10: 🥇 Won the National Gold Medal in the China International College Students' Innovation Competition (2024).
  • 2022.09: 🎉 Started my academic journey at Shanghai Jiao Tong University.

📖 Education

  • 2022.09 - 2026.06 (Expected), Shanghai Jiao Tong University, School of Computer Science, Software Engineering, Bachelor's Degree.
    Honors: National Scholarship, Meng Jia Chen Yue Scholarship, SJTU Class A Excellence Scholarship.
    Future: SJTU Computer Science Direct Master's Program (2026 Entry).

📝 Research Experience

Tech Report
Innovator-VL

Innovator-VL: A Multimodal Large Language Model for Scientific Discovery
Boxue Yang (Project Lead & First Author), et al.
[GitHub] [Project Page]

  • Led the development of a Multimodal LLM designed for science, open-sourcing both Instruct and Thinking versions.
  • Built a fully transparent, end-to-end reproducible training pipeline, achieving SOTA with less than 5 million high-quality scientific data samples.
  • Introduced the GSPO algorithm for the Thinking model, significantly improving inference efficiency. The project has received 100+ Stars.
ICML 2026 (Under Review)
EvoStreaming

EvoStreaming: Your Offline Video Model Is a Natively Streaming Assistant
Boxue Yang (First Author), et al.

  • Proposed a new paradigm to activate streaming video understanding capabilities without modifying the model architecture.
  • Constructed the first evaluation framework (RealStreamEval) that models streaming video understanding as frame-level multi-turn dialogue.
  • Designed a zero-human-annotation self-evolution pipeline (EvoStreaming), achieving a 10.8% performance boost on OVOBench with only 1k data.
Tech Report
AI4Service

AI for Service: Proactive Assistance with AI Glasses
Boxue Yang (Core Contributor & First Author), et al.

  • Proposed the "AI for Service" paradigm and built a proactive assistance system based on egocentric video streams.
  • Designed the Alpha-Service framework and an edge-side multi-agent collaborative architecture for resource-constrained AI glasses.
  • Realized the shift from "Human finding Service" to "AI finding Service," applied in scenarios like blackjack consulting and museum guiding.
ICML 2026 (Under Review)
Graph Learning

Robust Graph Representation Learning via Adaptive Spectral Contrast
Zhuolong Li, Boxue Yang, Haopeng Chen

  • Pioneered a spectral adversarial training framework for graph contrastive learning, optimizing objectives via a minimax game.
  • Designed dual-channel adaptive spectral filters and a novel Projected Gradient Attack (PGD), improving average classification accuracy by 2.82% on 9 benchmarks.
  • Significantly enhanced model robustness against poisoning attacks and noise perturbations.

🎖 Honors and Awards

Competitions

  • 2024.10 🥇 National Gold Medal, China International College Students' Innovation Competition (2024).
  • 2024.09 🥉 National Bronze Medal, The 11th "Challenge Cup" China Youth Innovation and Entrepreneurship Competition.
  • 2024.11 🥈 National Second Prize, National AI Application Scenario Innovation Challenge.

Academic & Personal Honors

  • 2025.12 🏅 National Scholarship (Top 1%, ¥10,000)
  • 2025.03 🏅 Meng Jia Chen Yue Scholarship (Top 1%, ¥10,000)
  • 2025.12 🏅 SJTU Class A Excellence Scholarship (Top 1%)
  • 2024.10 🎖 Merit Student of Shanghai Jiao Tong University
  • 2024.05 🎖 Outstanding League Member of Shanghai Jiao Tong University