Zhengyan Wan

I am a Master's student in the School of Statistics at East China Normal University, where I am fortunate to be advised by Prof. Fang Fang. Currently, I’m a visiting student in the Department of Statistics and Data Science at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), supervised by Prof. Qiang Sun. Previously, I received my Bachelor's degree in Statistics from the School of Mathematics at Jilin University. I am mainly interested in generative models, deep learning theory, and high-dimensional statistics.

Email  /  Github  /  Google Scholar

(* : Equal Contribution, ^† : Corresponding Author)

We propose two corrected samplers: time-corrected sampler and location-corrected sampler. We establish a one-step lower bound for the Euler sampler and non-asymptotic upper bounds for the proposed corrected samplers, showing that the location-corrected sampler achieves lower iteration complexity than the Euler sampler.

We establish non-asymptotic error bounds for distribution error in discrete flow models by taking approximation error into account. Both fast and slow convergence rates are derived with and without the lower-boundedness condition, respectively.

We derive the exact transition rate for the desired distribution given a learned discrete flow matching model, leading to guidance that only requires a single forward pass in each sampling step, significantly improving sampling efficiency.

Website credits go to Jon Barron!