👋 Hello, I’m Tianxiang
🎓 Research Interests
My research interest mainly centers on Time Series Analysis. I pay close attention to Foundation Models and Representation Learning. I try to understand the intrinsic nature of data through Statistical Physics and Information Theory. I also work on Uncertainty Theory to make models more reliable. I believe that a good model should handle data noise and concept drift well. I aim to solve real world problems with robust algorithms 🚀.
🌟 Experience & Future Goal
I also care about how to use these methods in specific domains. I have strong interests in Spatio-temporal Data. I have worked as a research intern at Griffith University and University of Macau. I also serve as a research fellow at PyPOTS because I enjoy coding and sharing my code 💻.
📖 Educations
- 2023.09 - 2026.06, Master’s Degree, University of Electronic Science and Technology of China, Computer Science. Mentor: Yong Deng
- 2019.09 - 2023.06, Bachelor‘s Degree, Southwest University (China), Software Engineering. Mentor: Fuyuan Xiao
💻 Internships
- 2025.07 - Now, PyPOTS, Research Fellow. Mentor: Qingsong Wen
- 2024.10 - Now, Griffith University (TrustAGI), Research Intern. Mentor: Ming Jin, Shirui Pan
- 2024.04 - 2025.06, PyPOTS, Community Scholars. Mentor: Qingsong Wen
- 2024.03 - Now, University of Macau, Research Intern. Mentor: Chi-Man Pun
- 2022.09 - Now, Chongqing University, Research Intern. Mentor: Fuyuan Xiao
Open Sources Project
Projects that have shaped how I think about evidence, time-series modeling, recursive self-improvement, and reproducible experimentation, listed from newest to oldest.
-
A lightweight recursive self-improvement loop for AGENTS.md-based agents.
- Drops into any compatible project without an SDK, runtime, or configuration layer.
- Runs diagnosis, independent evaluation, external research, implementation, and iterative closure as one loop.
- Keeps RSI-scoped memory plus separate web-research and blank-context evaluator agents.
-
An 85K-parameter pretrained TSFM built around fractal weight sharing.
- Forecasts without per-dataset fine-tuning using a release model with 85,001 parameters.
- Reuses one causal dilated-convolution block across a geometric ladder of temporal scales.
- Remains attention-free and fully convolutional, with bundled PyPI checkpoints and nine quantile outputs.
-
A reproducible experiment framework for deep learning and LLM workflows.
- Unifies configuration expansion, execution scheduling, result analysis, and one-call experiment pipelines.
- Supports sequential, process-pool, joblib, and dynamic execution with structured run artifacts and resume support.
- Includes CLI project templates and optional PyTorch, MLflow, W&B, and Excel integrations.
-
A Python package for evidence theory and uncertainty reasoning.
- Implements Dempster-Shafer, conjunctive, disjunctive, random-permutation-set, and orthogonal fusion rules.
- Supports belief, plausibility, commonality, pignistic transformation, Shafer discounting, and evidence moments.
- Ships as an installable package with API documentation and runnable examples.
-
A comprehensive code base for advanced deep time-series analysis.
- Covers long- and short-term forecasting, imputation, anomaly detection, and classification.
- Provides shared experiment and benchmark infrastructure for comparing advanced time-series models.
- Supports a broad model family, including Transformers, Mamba, and time-series foundation models.
-
A Python toolbox for machine learning on partially observed time series.
- Provides 50+ state-of-the-art models for imputation, classification, clustering, forecasting, anomaly detection, and cleaning.
- Handles incomplete, irregularly sampled multivariate time series with NaN missing values.
- Offers a unified Python interface, command-line tools, documentation, and reproducible examples.
📝 Publications
Total Citations:
😘 Selected Publications
Fracast-0: Fractal Weight Sharing for a Time Series Foundation Model with Only 85K Parameters
Tianxiang Zhan, Huanyao Zhang, and Yuanpeng He.
arXiv preprint arXiv:2609.32209 (2026).
AION: Next-Generation Tasks and Practical Harness for Time Series
Tianxiang Zhan, Xiaobao Song, Tong Guan, Shirui Pan, and Ming Jin.
arXiv preprint arXiv:2605.25045 (2026).
Ternary coding of maximum Deng entropy
Tianxiang Zhan, Yuanpeng He, and Yong Deng.
Fuzzy Sets and Systems (2026): 109913.
Time evidence fusion network: Multi-source view in long-term time series forecasting
Tianxiang Zhan, Yuanpeng He, Yong Deng, Zhen Li, Wenjie Du, and Qingsong Wen.
IEEE Transactions on Pattern Analysis and Machine Intelligence (2025).
Continuous Evolution Pool: Taming Recurring Concept Drift in Online Time Series Forecasting
Tianxiang Zhan, Ming Jin, Yuanpeng He, Yuxuan Liang, Yong Deng, and Shirui Pan.
arXiv preprint arXiv:2506.14790 (2025).
😁 Other Publications
End-to-End Learning for Partially-Observed Time Series with PyPOTS, Wenjie Du, Yiyuan Yang, Tianxiang Zhan, and Qingsong Wen.
Epistemic Uncertainty-Oriented Data Augmentation via Evidential Granulation, Binkai Liu, Qianli Zhou, Tianxiang Zhan, Neal N. Xiong, and Yong Deng.
Temperature-controlled power normalization of belief function, Xingyuan Chen, Tianxiang Zhan, and Yong Deng.
Evidential Prototype Learning for Semi-supervised Medical Image Segmentation, Yuanpeng He, Lijian Li, Tianxiang Zhan, Chi-Man Pun, Wenpin Jiao, and Zhi Jin.
Central moments of belief information, Xingyuan Chen, Duozi Lin, Tianxiang Zhan, Yong Deng.
Evaluating Evidential Reliability in Pattern Recognition Based on Intuitionistic Fuzzy Sets, Juntao Xu, Tianxiang Zhan, and Yong Deng.
Transfer Graph in the Context of Evidence Theory, Tianxiang Zhan, Qianli Zhou, Yuanpeng He, and Yong Deng.
Variance of belief information, Xingyuan Chen, Tianxiang Zhan, Guohui Zhou, and Yong Deng.
Learnable wsn deployment of evidential collaborative sensing model, Ruijie Liu, Tianxiang Zhan, Zhen Li, and Yong Deng.
Co-evidential fusion with information volume for semi-supervised medical image segmentation, Yuanpeng He, Lijian Li, Tianxiang Zhan, Chi-Man Pun, Wenpin Jiao, and Zhi Jin.
Residual feature-reutilization inception network, Yuanpeng He, Wenjie Song, Lijian Li, Tianxiang Zhan, and Wenpin Jiao.
Generalized information entropy and generalized information dimension, Tianxiang Zhan, Jiefeng Zhou, Zhen Li, and Yong Deng.
Generalized uncertainty-based evidential fusion with hybrid multi-head attention for weak-supervised temporal action localization, Yuanpeng He, Lijian Li, Tianxiang Zhan, Wenpin Jiao, and Chi-Man Pun.
Differential convolutional fuzzy time series forecasting, Tianxiang Zhan, Yuanpeng He, Yong Deng, and Zhen Li.
A fast evidential approach for stock forecasting, Tianxiang Zhan, and Fuyuan Xiao.
📷 Preprints
End-to-End Learning for Partially-Observed Time Series with PyPOTS, Wenjie Du, Yiyuan Yang, Tianxiang Zhan, and Qingsong Wen.
Co-Evidential Fusion with Information Volume for Medical Image Segmentation, Yuanpeng He, Lijian Li, Tianxiang Zhan, Chi-Man Pun, Wenpin Jiao, and Zhi Jin.
Residual Feature-Reutilization Inception Network for Image Classification, Yuanpeng He, Wenjie Song, Lijian Li, Tianxiang Zhan, and Wenpin Jiao.
Reliability Assessment of Information Sources Based on Random Permutation Set, Juntao Xu, Tianxiang Zhan, and Yong Deng.
Evaluating Evidential Reliability in Pattern Recognition Based on Intuitionistic Fuzzy Sets, Juntao Xu, Tianxiang Zhan, and Yong Deng.
Isopignistic Canonical Decomposition via Belief Evolution Network, Qianli Zhou, Tianxiang Zhan, and Yong Deng.
Random Graph Set and Evidence Pattern Reasoning Model, Tianxiang Zhan, Zhen Li, and Yong Deng.
Uncertainty Measurement of Basic Probability Assignment Integrity Based on Approximate Entropy in Evidence Theory, Tianxiang Zhan, Yuanpeng He, Hanwen Li, and Fuyuan Xiao.
Quantum Soft Likelihood Function Based on Ordered Weighted Average Operator, Tianxiang Zhan, Yuanpeng He, and Fuyuan Xiao.
A Matrix-Based Distance of Pythagorean Fuzzy Set and Its Application in Medical Diagnosis, Yuanpeng He, Lijian Li, and Tianxiang Zhan.
Construction Cost Index Forecasting: A Multi-Feature Fusion Approach, Tianxiang Zhan, Yuanpeng He, and Fuyuan Xiao.
DVS: Deep Visibility Series and Its Application in Construction Cost Index Forecasting, Tianxiang Zhan, Yuanpeng He, Hanwen Li, and Fuyuan Xiao.
🧑🏫 Mentor Roles
| Name | Mentorship Year | Previous Institution | Graduation Destination |
|---|---|---|---|
| Jingyou Wu | 2023 | University of Electronic Science and Technology of China | Zhejiang University |
| Juntao Xu | 2024 | University of Electronic Science and Technology of China | Tsinghua University |
| Mengyu Yang | 2024 | University of Electronic Science and Technology of China | - |
| Duozi Lin | 2024 | University of Electronic Science and Technology of China | Zhejiang University |
🤝 My collaborator
| Name | Institution |
|---|---|
| Yuanpeng He | Peking University |
| Lijian Li | University of Macau |
| Zhen Li | China Mobile |
| Jiefeng Zhou | University of South Florida |
| Qianli Zhou | Northwestern Polytechnical University |
| Mingxin Wang | GeWu-Lab, Renmin University of China |
| Xingyuan Chen | University of Electronic Science and Technology of China |
| Ruijie Liu | University of Electronic Science and Technology of China |
| Binkai Liu | University of Electronic Science and Technology of China |
| Xiaobao Song | Shenzhen University |
| Tong Guan | Zhejiang University |
| Yiyuan Yang | University of Oxford |