👋 About Me

Hi! I am a master's student in Computer Science and Technology at Shanghai Jiao Tong University (SJTU), where I received my bachelor's degree in 2025. I am currently a post-training intern at StepFun.

My research interests include agentic post-training, multimodal large language models, and efficient reasoning.

📖 Education

2025.09 – Present, M.S. Student

Computer Science and Technology

Shanghai Jiao Tong University, Shanghai, China

2021.09 – 2025.06, B.Eng.

Computer Science and Technology

Shanghai Jiao Tong University, Shanghai, China

🔥 News

  • 🎉 Two first-author papers accepted to NeurIPS 2026!
  • 🎉 Two papers accepted to EMNLP 2026: one Main and one Findings!
  • 🎉 One paper accepted to ICML 2026!
  • 🎉 One first-author paper accepted to ACL 2026 (Main)!
  • 🎉 One paper accepted to ICLR 2026!
  • 🎉 One paper accepted to WWW 2026!
  • 🎉 Our first-author paper D³ToM accepted to AAAI 2026!

📝 Publications

My name is in bold. * denotes equal contribution.

Accepted Papers 9 papers

Overview of rendered compression and discrete latent reasoning.
NeurIPS 2026First author

Why Struggle with Continuous Latents? Interpretable Discrete Latent Reasoning via Rendered Compression

Compresses rendered reasoning traces into discrete latent tokens for supervised and reinforcement learning.

Shuochen Chang, Qingyang Liu, Shaobo Wang, Bingjie Gao, Qianli Ma, Haonan Zhao, Yibo Miao, Yulin Sun, Zelin Peng, Jiangtong Li, Li Niu

PaperCode
NeurIPS 2026First author

From Label Priors to Task Evidence: Long-Video Frame Selection via Bayesian GRPO

Uses Bayesian GRPO to select task-relevant frames rather than rely on label priors in long videos.

Shuochen Chang, Bingjie Gao, Qingyang Liu, Qianli Ma, Yibo Miao, Haonan Zhao, Zhaohe Liao, Xiaofeng Zhang, Jiangtong Li, Li Niu

Latent reasoning analysis from the paper repository.
ACL 2026 MainFirst author

Unlocking the Black Box of Latent Reasoning: An Interpretability-Guided Approach to Intervention

Uses interpretability probes to guide training-free interventions that improve latent reasoning accuracy.

Shuochen Chang, Tong Bai, Xiaofeng Zhang, Qianli Ma, Qingyang Liu, Zhaohe Liao, Yibo Miao, Li Niu

PaperCode
D³ToM decider-guided token merging architecture.
AAAI 2026First author

D³ToM: Decider-Guided Dynamic Token Merging for Accelerating Diffusion MLLMs

Merges visual tokens using decider-token importance to accelerate diffusion MLLMs without retraining.

Shuochen Chang, Xiaofeng Zhang, Qingyang Liu, Li Niu

PaperCode
Comparison of answer agreement and visually grounded evidence.
EMNLP 2026 FindingsCo-first author

Seeing Before Agreeing: Aligning Multi-Agent Consensus with Visual Evidence

Verifies and revises shared visual evidence to guide multi-agent consensus in visual question answering.

Yuhan Wang*, Shuochen Chang*, Yalin Feng, Dongsheng Ma, Yuanzi Li, Zhengren Wang, Yinglong Yang, Yufei Chen, Yikang Wang, Shaoxu Sun, Wentao Zhang

Paper
EMNLP 2026 MainThird author

Residual-Memory-aware Token Merging for Accelerating Diffusion MLLMs

Uses residual-memory-aware token merging to accelerate diffusion MLLMs.

Yifei Bao, Jingxing Zhong, Shuochen Chang, Yifan Wang, Xingyou Fang, Pengfei Guo, Xiaofeng Zhang

Interleaved visual reasoning tasks.
ICML 2026Seventh author

Breaking Dual Bottlenecks: Evolving Unified Multimodal Models into Self-Adaptive Interleaved Visual Reasoners

Trains unified multimodal models to generate, reflect, and plan with interleaved visual reasoning.

Qingyang Liu, Bingjie Gao, Canmiao Fu, Zhipeng Huang, Chen Li, Feng Wang, Shuochen Chang, Shaobo Wang, Yali Wang, Keming Ye, Jiangtong Li, Li Niu

PaperCode
Context information flow and the repetition problem in diffusion MLLMs.
ICLR 2026Third author

Context Tokens are Anchors: Understanding the Repetition Curse in dMLLMs from an Information Flow Perspective

Reduces cache-induced repetition by preserving context-token information flow and adjusting decoding confidence.

Qiyan Zhao*, Xiaofeng Zhang*, Shuochen Chang, Qianyu Chen, Xiaosong Yuan, Xuhang Chen, Luoqi Liu, Jiajun Zhang, Xu-Yao Zhang, Da-Han Wang

PaperCode
Short-drama quality assessment task illustration.
WWW 2026Sixth author

Bridging Visual Dynamics and Narrative Reasoning: Multimodal Large Language Models for Short Drama Quality Assessment

Combines visual and narrative reasoning with preference-aligned rewards to assess short-drama quality.

Qingyang Liu*, Jiangtong Li*, Zelin Peng, Shaobo Wang, Zhaohe Liao, Shuochen Chang, Bingjie Gao, Haonan Zhao, Mu Liu, Jidong Jiang, Li Niu

PaperCode

💻 Experience

  • 2026.06 – Present, StepFun, Post-Training Intern.
  • 2025.11 – 2026.05, TikTok, ByteDance, Research Intern.
  • 2025.04 – 2025.10, 4Paradigm, Research Intern.

🎖 Honors & Awards

  • Undergraduate Academic Excellence Scholarship2023–2024
  • Zhiyuan Honors Scholarship2021–2024