Wei Deng

I am a researcher at Morgan Stanley. My interest is to study GPU-friendly sampling, diffusion, filtering, and transport methods. The applications include, but not limited to generative models and time series.

I got my Ph.D. at Purdue University in 2021, where I was advised by Prof. Guang Lin and Faming Liang.

Feel free to contact me at: firstnamelastname056@gmail.com


Apr, 2024. Variational Schrödinger Diffusion Models is accepted by ICML’24! Schrödinger diffusion is scalable now by linearizing the forward scores. Simulation-free property is all we need for scalability.

Apr, 2024. Reflected Schrödinger Bridge is accepted by UAI’24 as Oral (<3% acceptance rate). Reflected Schrödinger has the linear convergence of couplings via entropic optimal transport on bounded domains.

Apr, 2024. 2 ICML, 2 UAI (+1 Oral) and 1 AISTAT are accpeted!

Feb, 2024. 1 JCGS (revision)

Feb, 2024. Talk at Statistics Colloquium at UConn

Dec, 2023. Participate Transport, Diffusion, and Sampling Workshop at Flatiron Institute.

Nov, 2023. Talk at Financial Mathematics Seminar at FSU

Apr, 2023. One ICML on diffusion Schrodinger bridge with transformers is accepted!

Dec, 2022. The Contour Sampler is implemented in BlackJAX!

Feb, 2022. Talk at Opt and ML Seminar at HKU

Oct, 2021. I have defended my thesis!

Oct, 2020. Talk at ML + X Seminar at Brown University


Reviewers: ICML, NeurIPS, ICLR, AISTAT, AAAI, IJCAI, Machine Learning Journal.

La pensée n’est qu’un écliar au milieu d’une longue nuit. Mais c’est cet éclair qui est tout.