Mingjia (Samuel Jayden) Shi’s Homepage

• I am a Ph.D. student in the VAST Lab at the University of Virginia, advised by Jundong Li.
• I was an AI R&D Intern at Nokia in Sunnyvale, California, from May to August 2026, where I worked on long-context compression and robustness.
• I interned at HoumoAI until July 2025, working on resource-efficient AI for real-world and industrial applications.
• I spent two years as a research intern at NUS HPC-AI Lab, where I worked on resource-efficient AI and generative models.
• From September 2021 to June 2024, I completed my master’s degree in Artificial Intelligence (Data Science) at Sichuan University, where I had also earned my bachelor’s degree. I was supervised by Prof. Jiancheng Lv and studied federated learning, decentralized data analysis, and large-scale learning systems.

• [Sep. 26] My first-author paper SAP was accepted by NeurIPS 2026, and two co-authored papers, DD-Ranking and MemSuit, were accepted by NeurIPS 2026 (Oral) and EMNLP 2026, respectively.
• [Jul. 26] One first-author paper was accepted by ECCV 2026, and one collaborative paper was accepted by ICML 2026.
• [Aug. 25] I began my Ph.D. in Computer Engineering at UVA. I welcome research collaborations, joint proposals, and other academic partnerships.

I am currently based at the University of Virginia in Charlottesville. My research focuses on efficient AI, with an emphasis on making modern AI systems more capable, scalable, and resource-efficient. I welcome research collaborations and joint proposals in these areas.

My primary research focus is Efficient AI. I study efficiency across the AI lifecycle, including efficient training and inference, model and data efficiency, multimodal systems, and AI-native approaches that jointly design models, algorithms, and systems around real computational constraints.

• 🎓 Overall background. My background spans mathematics and computer science, optimization and generalization, systems, control and information theory, generative modeling, multimodal learning, and AI-native efficiency.
• 🤝 Collaboration. I welcome collaborations on efficient AI, particularly projects connecting learning algorithms, multimodal models, and systems-level efficiency.
• 🌟 Broader interests. Beyond technical AI research, I remain interested in fundamental questions across physics and the natural sciences.

📅 2026:

1. NeurIPS 2026 Saliency-Aware Multi-Route Thinking: Grounding and Reasoning on Vision-Language Agents. M. Shi, Y. He, Y. Zhu, J. Li (preprint, details)
2. ECCV 2026 Dive into the Implicit Biases of Low-rank Vision-language Alignment. M. Shi, S. Wang, X. Wang, S. Zhou, K. Wang, T. Fu, C. Zhao, A. Su, P. Jiang, M. Wu (paper)
3. NeurIPS 2026 (Oral) DD-Ranking: Rethinking the Evaluation of Dataset Distillation. Z. Li, X. Zhong, S. Khaki, Z. Liang, Y. Zhou, M. Shi, Z. Wang, X. Zhao, W. Zhao, Z. Qin, M. Wu, P. Zhou, H. Wang, D. J. Zhang, J. Liu, S. Wang, D. Liu, L. Zhang, G. Li, K. Wang, Z. Zhu, Z. Ma, J. T. Zhou, J. Lv, Y. Jin, P. Wang, K. Zhang, L. Lyu, Y. Huang, Z. Akata, Z. Deng, X. Wu, G. Cazenavette, Y. Shang, J. Cui, J. Gu, Q. Zheng, H. Ye, S. Wang, X. Wang, Y. Yan, A. Yao, M. Z. Shou, T. Chen, H. Bilen, B. Mirzasoleiman, M. Kellis, K. N. Plataniotis, Z. Wang, B. Zhao, Y. You, K. Wang (paper)
4. EMNLP 2026 Grounding Memory Summarization in Utility Intent. Z. Lei, M. Shi, X. Fu, H. He, Q. R. Wang, J. Li (paper, details)
5. ICML 2026 IAPO: Information-Aware Policy Optimization for Token-Efficient Reasoning. Y. He, Y. Zhu, M. Shi, W. Zheng, L. Su, X. Wang, Q. Guo, J. Li (paper, code)

📅 2025:

1. Arxiv 2025 Make Optimization Once and for All with Fine-grained Guidance. Arxiv. M. Shi, R. Lin, X. Chen, Y. Zhou, Z. Ding, P. Li, T. Wang, K. Wang, Z. Wang, J. Zhang, T. Chen. (paper)
2. CVPR 2025 A Closer Look at Time Steps is Worthy of Triple Speed-Up for Diffusion Model Training. K. Wang*, M. Shi*, Y. Zhou, Z. Li, Z. Yuan, Y. Shang, X. Peng, H. Zhang, Y. You (paper, code)
3. NeurIPS 2025 Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights. Z. Liang, D. Tang, Y. Zhou, X. Zhao, M. Shi, W. Zhao, Z. Li, P. Wang, K. Schürholt, D. Borth, M. M. Bronstein, Y. You, Z. Wang, K. Wang (paper)
4. NeurIPS 2025 Pruning-Robust Mamba with Asymmetric Multi-Scale Scanning Paths. J. Lv, Y. Zhou, M. Shi, Z. Liang, P. Zhang, X. Peng, W. Zhao, Z. Zhu, J. Lv, Q. Ye, K. Wang (paper)
5. NeurIPS 2025 Tackling Feature-Classifier Mismatch in Federated Learning via Prompt-Driven Feature Transformation. X. Wu, J. Niu, X. Liu, M. Shi, G. Zhu, S. Tang (paper)
6. NeurIPS 2025 REPA Works Until It Does not: Early-Stopped, Holistic Alignment Supercharges Diffusion Training. Z. Wang, W. Zhao, Y. Zhou, Z. Li, Z. Liang, M. Shi, X. Zhao, P. Zhou, K. Zhang, Z. Wang, K. Wang, Y. You (paper)
7. CVPR 2025 Ferret: An Efficient Online Continual Learning Framework under Varying Memory Constraints. Y. Zhou, Y. Tian, J. Lv, M. Shi, Y. Li, Q. Ye, S. Zhang, J. Lv (paper)
8. TNNLS E-3SFC: Communication-Efficient Federated Learning With Double-Way Features Synthesizing. Y. Zhou, Y. Tian, M. Shi, Y. Li, Y. Sun, Q. Ye, J. Lv (paper)
9. ACL 2025 Findings GSQ-Tuning: Group-Shared Exponents Integer in Fully Quantized Training for LLMs On-Device Fine-tuning. S. Zhou, S. Wang, Z. Yuan*, M. Shi, Y. Shang, D. Yang (paper)
10. IJCNN 2025 FedSH: Tackling Staleness By Scheduling High-order Approximation in Asynchronous Federated Learning. H. Gao, M. Shi, Y. Zhou, D. Xiong, Q. Ye, J. Lv (paper)

📅 Early Selected:

Released Pre-Print

1. Arxiv 2024 Faster Vision Mamba is Rebuilt in Minutes via Merged Token Re-training. M. Shi*, Y. Zhou*, R. Yu, Z. Li, Z. Liang, X. Zhao, X. Peng, T Rajpurohit, R. Vedantam, W. Zhao, K. Wang, Y. You. (paper, code, page)

Conference

1. ICASSP 2024 Federated CINN Clustering for Accurate Clustered Federated Learning. Y. Zhou, M. Shi, Y. Tian, Y. Li, Q. Ye, J. Lv (paper)
2. NeurIPS 2023 PRIOR: Personalized Prior for Reactivating the Information Overlooked in Federated Learning. M. Shi, Y. Zhou, K. Wang, H. Zhang, S. Huang, Q. Ye, J. Lv (paper, code)
3. ICONIP 2023 Unconstrained Feature Model and Its General Geometric Patterns in Federated Learning: Local Subspace Minority Collapse. M. Shi, Y. Zhou, Q. Ye, J. Lv (paper)
4. ICCV 2023 Communication-efficient Federated Learning with Single-Step Synthetic Features Compressor for Faster Convergence. Y. Zhou, M. Shi, Y. Li, Y. Sun, Q. Ye, J. Lv (paper)

Journal

1. InfoSci DeFTA: A Plug-and-Play Peer-to-Peer Decentralized Federated Learning Framework. Y. Zhou, M. Shi, Y. Tian, Q. Ye, J. Lv (paper)
2. Trans.ETCI DLB: a dynamic load balance strategy for distributed training of deep neural networks. Q. Ye, Y. Zhou, M. Shi, Y. Sun, J. Lv (paper)
3. JoSc FLSGD: free local SGD with parallel synchronization. Q. Ye, Y. Zhou, M. Shi, J. Lv (paper)

Please see the full list in Google Scholar

Selected conference reviewing: NeurIPS, ICML, and ICLR; CVPR, ICCV, and ECCV; ACL Rolling Review (ARR), serving ACL, EMNLP, NAACL, and other ARR venues; ACM Multimedia; IJCNN, ICPP, ICASSP, ICONIP, etc.

Journal reviewing: TNNLS, TCSVT, TPDS, TKDE, Neurocomputing, and others.

📷 Photograph, ⛺ Travel, 🎵 Music, 🏸 Badminton, ⚽ Football (Soccer), ♟️ Chess and other Table-top Games, 🎱 Billiards, 🎮 Games, 🔬 Learning about interesting topics.