2026.10
Awarded the National Scholarship.
SJTU × Beijing Zhongguancun Academy / Computer Science
Wei Li Machine Learning Research
PhD Student in Computer Science at Shanghai Jiao Tong University × Beijing Zhongguancun Academy (2027-), working on machine learning for structured and temporal data. Current work spans generative modeling, forecasting, clustering, representation learning, and uncertainty-aware modeling. Undergraduate education: Computer Science and Technology, Shanghai University (2023-2027).
2026.09
SDFlow accepted for Poster Presentation at NeurIPS 2026 (CCF-A).
2026.08
The APCL (KDD 2026, CCF-A) blog post is now live.
2026.06
APCL scheduled for Poster Presentation at KDD 2026 (CCF-A).
Featured Work
Selected papers at a glance.
A compact view of representative work. The full publication list includes abstracts, BibTeX, figures, and code.
SDFlow: Similarity-Driven Flow Matching for Time Series Generation
SDFlow introduces a similarity-driven, non-autoregressive flow-matching framework for time-series generation in high-dimensional discrete latent spaces. By learning a low-rank subspace and initializing generation with similarity-guided manifold anchors, it aligns the generative process with the geometry of real temporal data and supports parallel synthesis.
Details & BibTeX →
Adaptive Prototypical Contrastive Learning for Time Series Clustering
Addresses time-series clustering with an unknown number of clusters by combining hierarchical prototypes, contrastive learning, and the MDL principle to jointly learn representations and cluster cardinality.
Details & BibTeX →
ClusterPatchTST: Uncertainty-Aware Causal Clustering for Heterogeneous Time Series Forecasting
Uses causal clustering to capture cross-series structural heterogeneity and uncertainty modeling to improve robust and reliable forecasting.
Details & BibTeX →
EnergyPatchTST: Multi-scale Time Series Transformers with Uncertainty Estimation for Energy Forecasting
Combines multiscale Patch Transformers with uncertainty estimation for energy forecasting; the platform received a software copyright and 300+ GitHub stars.
Details & BibTeX →