Shilin Ma

Hi! I am a PhD student at Tsinghua University, advised by Prof. Yansong Tang.

Previously, I received my B.Eng. degree in Electronic Engineering from Tsinghua University in 2025.

My research focuses on physical agents and long-horizon embodied tasks, with an emphasis on agent memory, planning, and closed-loop control.

Email / GitHub / OpenReview

Education

	Tsinghua University 2025.09 - Present PhD student, Shenzhen International Graduate School Advisor: Prof. Yansong Tang

	Tsinghua University 2021.09 - 2025.06 B.Eng. in Electronic Engineering Department of Electronic Engineering

Industrial Experience

Tencent Robotics X 2026.05 - Present

Research Intern
Research on embodied agents for long-horizon tasks

Recent Publications

(* Equal contribution, ^† Corresponding author)

	SAFE-Pruner: Semantic Attention-Guided Future-Aware Token Pruning for Efficient Vision-Language-Action Manipulation Shilin Ma, Chubin Zhang, Changyuan Wang, Yuji Wang, Yue Wu, Zixuan Wang, Jingqi Tian, Zheng Zhu, Yansong Tang^† European Conference on Computer Vision (ECCV), 2026 [arXiv] [Paper] [Code] [Project Page] We propose a training-free, future-aware visual token pruning framework for efficient VLA manipulation.
	VARestorer: One-Step VAR Distillation for Real-World Image Super-Resolution Yixuan Zhu, Shilin Ma, Haolin Wang, Ao Li, Yanzhe Jing, Yansong Tang^†, Lei Chen, Jiwen Lu, Jie Zhou International Conference on Learning Representations (ICLR), 2026 [arXiv] [Paper] [Code] [Project Page] We distill a visual autoregressive model into a one-step real-world image super-resolution model.
	FADE: Frequency-Aware Diffusion Model Factorization for Video Editing Yixuan Zhu, Haolin Wang, Shilin Ma, Wenliang Zhao, Yansong Tang^†, Lei Chen, Jie Zhou^† IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025* [arXiv] [Paper] [Code] We factorize diffusion features by frequency to balance structural consistency and appearance change in video editing.
	Uncertainty-Aware World Model for Aerial Image-Goal Navigation Deyi Zhu, Haoyu Fan, Yinan Zhu, Weichen Zhang, Shilin Ma, Xinlei Chen, Yansong Tang Preprint, 2026 [arXiv] [Paper] [Project Page] We formulate trajectory scoring as conditional out-of-distribution detection for robust aerial navigation under future-state uncertainty.
	AdaThinkV: Adaptive Thinking for Token-Efficient Video Reasoning Jingqi Tian, Haoji Zhang, Lin Chen, Hongbo Jin, Haonan Xu, Tianrui Zhu, Xingming Shui, Shilin Ma, Wenjing Yang, Yansong Tang^† Preprint, 2026* [arXiv] [Paper] [Code] [Project Page] We propose an adaptive video-reasoning framework that learns when explicit reasoning is worth its generation cost.
	MorphoQuant: Modality-Aware Quantization for Omni-modal Large Language Models Yue Wu, Changyuan Wang, Zixuan Wang, Shilin Ma, Yansong Tang Preprint, 2026 [arXiv] [Paper] We develop a modality-aware post-training quantization framework that preserves cross-modal structure at 4-bit precision.