JJJingyan Jiang
中文
Open to academic & industry collaboration

👋 I am Jingyan Jiang, an artificial intelligence researcher studying how intelligent systems adapt, remember, and reason in open, dynamic environments. My current work focuses on test-time and continual adaptation and multimodal embodied agents.

🎓 I conducted postdoctoral research at Tsinghua Shenzhen International Graduate School under Profs. Zhi Wang and Shu-Tao Xia. Earlier, I was a visiting Ph.D. researcher there advised by Prof. Zhi Wang, and received my Ph.D. in Computer Science from Jilin University under Prof. Liang Hu.

🔬 I explore reliable adaptation under distribution shifts, long-term knowledge retention, and memory-augmented spatio-temporal reasoning. My work has appeared at venues including CVPR, ECCV, NeurIPS, AAAI, ACM MM, ACL, and IEEE RTSS.

01

Current Research

I focus on two complementary questions: how models adapt at deployment time, and how embodied agents remember and reason over long horizons.

Test-Time & Continual Adaptation

Reliable adaptation under changing test distributions.

  • Robust adaptation for vision-language models and dynamic shifts.
  • Continual knowledge retention without retraining.

Multimodal & Embodied Agents

Memory and reasoning for agents in dynamic worlds.

  • Long-horizon robotic memory and manipulation.
  • Multimodal 4D evidence tracking and reasoning.
02

News

  • 🎉 RoboStream · ECCV 2026; CurvSpec · ACM MM 2026.

  • 🎉 Neural Collapse in Test-Time Adaptation · CVPR 2026.

  • 🎉 MoETTA · AAAI 2026; RLVR reasoning · Findings of ACL 2026.

  • 🏆 Received the Second Prize of the China Federation of Logistics & Purchasing Science and Technology Award.

  • 🎉 Recent work appeared at NeurIPS 2025 and CVPR 2025.

  • 🏆 Received the Special Prize of the Guangdong Higher Education Teaching Achievement Award.

03

Selected Publications

Selected work in embodied intelligence and test-time adaptation. * first author / equal contribution · † corresponding author

PREPRINT 2026ChainVLA architecture

ChainVLA: Chaining Vision-Language-Action Queries through a Unified Execution State for Long-Horizon Manipulation

Yuzhi Huang*, Weijue Bu*, Ziyi Xiong, Jie Wu, Fanding Huang, Jingyan Jiang†, Zhi Wang†

A 1.2B VLA policy that carries task progress and unfinished motion across queries for long-horizon manipulation.

Vision-Language-ActionProjectPaper
ECCV 2026RoboStream framework overview

RoboStream: Weaving Spatio-Temporal Reasoning with Memory in Vision-Language Models for Robotics

Yuzhi Huang*, Jie Wu*, Weijue Bu*, Ziyi Xiong, Gaoyang Jiang, Ye Li, Kangye Ji, Shuzhao Xie, Yue Huang, Chenglei Wu, Jingyan Jiang†, Zhi Wang†

Persistent spatio-temporal tokens and causal memory enable long-horizon robotic reasoning and manipulation.

Embodied IntelligenceProjectPaperCode
CVPR 2026NCTTA framework overview

Neural Collapse in Test-Time Adaptation

Xiao Chen*, Zhongjing Du, Jiazhen Huang, Xu Jiang, Li Lu, Jingyan Jiang†, Zhi Wang†

A neural-collapse perspective that realigns shifted features with classifier geometry through hybrid targets.

Test-Time AdaptationPaperCode
PREPRINT 2026ZAEC diagnostic analysis

Respect Your Zero-Shot Uncertainty: Conservative Calibration for Test-Time-Adapted Vision-Language Models

Jingyan Jiang*, Yaru Sun*, Xiao Chen, Jiazhen Huang, Caiting Li, Zhijian He, Yin Chen†, Pingting Hao†

A label-free post-hoc method that anchors adapted confidence to each sample’s zero-shot uncertainty.

TTA CalibrationPaper
NEURIPS 2025FIND framework overview

Feature-Based Instance Neighbor Discovery: Advanced Stable Test-Time Adaptation in Dynamic World

Qinting Jiang*, Chuyang Ye, Dongyan Wei, Bingli Wang, Yuan Xue, Jingyan Jiang†, Zhi Wang†

A layer-wise divide-and-conquer normalization framework for stable adaptation to mixed dynamic distributions.

Dynamic TTAPaperCode
CVPR 2025COSMIC framework overview

COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation

Fanding Huang*, Jingyan Jiang*, Qinting Jiang, Hebei Li, Faisal Nadeem Khan†, Zhi Wang†

Robust CLIP adaptation through dual semantic graphs and clique-guided hyper-class reasoning.

CLIP AdaptationPaperCode
IEEE RTSS 2021modeldata

Joint Model and Data Adaptation for Cloud Inference Serving

Jingyan Jiang*, Ziyue Luo, Chenghao Hu, Zhaoliang He, Zhi Wang, Shu-Tao Xia, Chuan Wu†

Jointly adapts model configuration and input data to balance inference accuracy and latency under changing system conditions.

Edge IntelligencePaper
Curated publication archiveBrowse 38 selected computer-science papers→
04

Honors & Awards

  • 2026 · China Federation of Logistics & Purchasing Science and Technology Award — Second Prize.
  • 2025 · Guangdong Higher Education Teaching Achievement Award — Special Prize.
05

Academic Service

  • Area Chair: IEEE ICME.
  • Reviewer: IEEE TCSVT, NeurIPS, CVPR, ECCV, ACM MM, AAAI, ICASSP, and ICME.
✉

Interested in test-time adaptation or multimodal embodied agents? Let’s start a conversation.

06

Visitors

Global visitor mapApproximate, aggregate locations
—Unique visitors
—Page views

Thank you for visiting from around the world. Counts begin accumulating after the site is published.