About Me

I’m Junlong Tong, a Ph.D. candidate in Computer Science at Shanghai Jiao Tong University and Eastern Institute of Technology (supervised by Dr. Xiaoyu Shen).

My research focuses on multimodal foundation models for dynamic real-world environments, with the goal of enabling models to perceive, reason, and interact continuously and proactively in real time. In particular, I focus on two complementary directions:

• Streaming LLMs/MLLMs, which study real-time multimodal perception, concurrent reasoning, and proactive interaction over continuous, long-horizon multimodal streams.
• Efficient LLMs/MLLMs, which investigate data/token/KV cache compression, layer pruning, and efficient inference for scalable deployment.

More broadly, I aim to build multimodal foundation models that are both capable of operating in continuous real-world settings and efficient enough for practical deployment at scale, and serve as the foundation for multimodal agents. I also have research experience in LLMs for time-series modeling, which complements my broader interest in temporal modeling over continuous data.

💡 Seeking Research Internship Opportunities & Collaborations: I am actively seeking research internship opportunities in LLMs/MLLMs and am always open to academic collaborations. Feel free to reach out to me via jl-tong@sjtu.edu.cn!

News ✨

[2026.04] The survey of streaming LLMs has been accepted by ACL 2026 Findings.🎉
[2026.02] Three papsers are accepted by CVPR 2026.🎉
[2026.01] Two papers are accepted by ICLR 2026.🎉
[2025.08] One paper is accepted by EMNLP 2025.🎉
[2025.06] One paper is accepted by ICCV 2025.🎉
[2025.05] One paper is accepted by ACL 2025 Findings.🎉
[2025.05] One paper is accepted by ICML 2025.🎉

Publications

A full list of publications can be found on Google Scholar.
(* Equal Contribution, ^† Corresponding author)

• Streaming LLMs/MLLMs (Demo)

1. From Static Inference to Dynamic Interaction: A Survey of Streaming Large Language Models. [PDF] [Repository]
   Junlong Tong, Zilong Wang, YuJie Ren, Peiran Yin, Hao Wu, Wei Zhang, Xiaoyu Shen^†.
   Findings of ACL 2026.
   
2. StreamingThinker: Large Language Models Can Think While Reading. [PDF] [Code] [Project]
   Junlong Tong, Yingqi Fan, Anhao Zhao, Yunpu Ma, Xiaoyu Shen^†.
   ICLR 2026.
   
3. Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models. [PDF] [Code] [Project]
   Jialiang Zhang*, Junlong Tong*, Junyan Lin, Hao Wu, Yunpu Ma, Xiaoyu Shen^†. (* Equal Contribution)
   CVPR 2026.
   
4. LLM as Effective Streaming Processor: Bridging Streaming-Batch Mismatches with Group Position Encoding. [PDF] [Code]
   Junlong Tong, Jinlan Fu, Zixuan Lin, Yingqi Fan, Anhao Zhao, Hui Su, Xiaoyu Shen^†.
   Findings of ACL 2025.
   
5. ProactiveLLM: Learning Active Interaction for Streaming Large Language Models.
   Junlong Tong, Yao Zhang, Anhao Zhao, Yingqi Fan, Yunpu Ma, Xiaoyu Shen^†.
   Under review
   
6. Speak While Watching: Unleashing TRUE Real-Time Video Understanding Capability of Multimodal Large Language Models. [PDF][Code]
   Junyan Lin*, Junlong Tong*, Hao Wu, Jialiang Zhang, Jinming Liu, Xin Jin, Xiaoyu Shen^†.
   Under review
   

• Efficient LLMs/MLLMs

1. Context Guided Transformer Entropy Modeling for Video Compression. [PDF]
   Junlong Tong, Wei Zhang, Yaohui Jin, Xiaoyu Shen^†.
   ICCV 2025.
   
2. What Do Visual Tokens Really Encode? Uncovering Sparsity and Redundancy in Multimodal Large Language Models. [PDF][Code]
   Yingqi Fan, Junlong Tong, Anhao Zhao, Xiaoyu Shen^†.
   CVPR 2026. (Highlight)
   
3. HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit [PDF][Code]
   Hao Wu, Yingqi Fan, Jinyang Dai, Junlong Tong, Yunpu Ma, Xiaoyu Shen^†
   ICLR 2026.
   
4. VisiPruner: Decoding Discontinuous Cross-Modal Dynamics for Efficient Multimodal LLMs. [PDF]
   Yingqi Fan, Anhao Zhao, Jinlan Fu, Junlong Tong, Hui Su, Yijie Pan, Wei Zhang, Xiaoyu Shen^†.
   EMNLP 2025.
   
5. SkipGPT: Each Token is One of a Kind. [PDF] [Code]
   Anhao Zhao, Fanghua Ye, Yingqi Fan, Junlong Tong, Jing Xiong, Zhiwei Fei, Hui Su, Xiaoyu Shen^†.
   ICML 2025.
   
6. From Data to Model: A Survey of the Compression Lifecycle in MLLMs. [PDF] [Repository]
   Hao Wu*, Junlong Tong*, Xudong Wang, Yang Tan, Changyu Zeng, Anastasia Antsiferova, Xiaoyu Shen^†.
   Under review
   

• LLM for Sequence Modeling

1. Rethinking the Role of LLMs in Time Series Forecasting. [PDF] [Code]
   Xin Qiu*, Junlong Tong*, Yirong Sun, Yunpu Ma, Wei Zhang, Xiaoyu Shen^†.
   Under review
   
2. The Few Govern the Many: Unveiling Few-Layer Dominance for Time Series Models. [PDF] [Code]
   Xin Qiu*, Junlong Tong*, Yirong Sun, Yunpu Ma, Xiaoyu Shen^†.
   Under review
   
3. Probabilistic Decomposition Transformer for Time Series Forecasting. [PDF] [Code]
   Junlong Tong, Liping Xie, Kanjian Zhang.
   SIAM International Conference on Data Mining (SDM 2023).
   
4. Enhancing Time Series Forecasting: A Hierarchical Transformer with Probabilistic Decomposition Representation. [PDF]
   Junlong Tong, Liping Xie, Wankou Yang, Kanjian Zhang, Junsheng Zhao.
   Information Sciences 2023.
   
5. Hourly Solar Irradiance Forecasting Based on Encoder–decoder Model Using Series Decomposition and Dynamic Error Compensation. [PDF]
   Junlong Tong, Liping Xie, Shixiong Fang, Wankou Yang, Kanjian Zhang.
   Energy Conversion and Management 2022.
   

Educations

• Ph.D. candidate in Computer Science at Shanghai Jiao Tong University (2023 - present), Shanghai, China
• M.Eng in Artificial Intelligence at Southeast University (2020 - 2023), Nanjing, China
• B.Eng in Electrical Engineering at Shanghai University of Electric Power (2016 - 2020), Shanghai, China

Honors and Awards

• Outstanding Master’s Thesis Award of Jiangsu Province, 2023
• SDM23 Student Travel Awards, 2023

Academic Services

• Conference Reviewer: ICML, NeurIPS, ACL, CVPR, ECCV, AAAI
• Journal Reviewer: TCVST, TNNLS

Contact

Email: jl-tong@sjtu.edu.cn