Hey, this is Ganchao Wei (魏赣超 in Chinese), a stats / ML player (or AI, if you want) in New York, NY, mostly motivated by problems in biological science, especially neuroscience.

I used to play more with Bayesian and latent variable models for spatiotemporal data, and recently I’ve been into deep generative models (diffusion- / flow-based) and causal stuff.

Some recent ideas on flow stuff

For count data

• Flow matching directly on count space.
  Ganchao Wei and John Pearson. Flow Matching for Count Data. NeurIPS, 2026.

• Finite-time stochastic transitions for one- / few-step count generation.
  Ganchao Wei. Stochastic Flow Map for Count Data. arXiv preprint arXiv:2609.23290, 2026. [in submission]

For time series

• GP stochastic paths for flow matching with correlated / time-series data.
  Ganchao Wei and Li Ma. Stream-level Flow Matching with Gaussian Processes. ICML, 2025.

• Flexible low-D dynamics, with applications to neuroscience.
  Ganchao Wei, Daniela de Albuquerque, Miles Martinez, Shiyang Pan, and John Pearson. Dynamic Compression Flows for Neuroscience Data. ICML, 2026.

For causal discovery / representation learning

• Using diffusion / flow models to sample causal graphs, for high-D graphs, uncertainty, Markov equivalence classes, etc.

  Coming soon… hopefully 😅

---

If you have an idea you’d like to chat about, please shoot me an email at weiganchao[at]gmail[dot]com!