Michael Zhu

Papers

Total Citations

H-Index

About

Michael Zhu is a researcher at the forefront of differentiable inference and probabilistic machine learning, with a particular focus on enabling end-to-end learning in sample-based state estimation. His most cited work, "Towards Differentiable Resampling" (2020, 21 citations), tackles a fundamental bottleneck in differentiable particle filters: the non-differentiability of the resampling step. By proposing novel approaches to make resampling compatible with gradient-based optimization, Zhu has opened new pathways for integrating classic Bayesian filtering methods into deep learning pipelines. This contribution is critical for applications in robotics, tracking, and sequential decision-making where uncertainty must be learned from data. While his citation count reflects the emerging nature of this specialized field, Zhu’s work is highly influential among researchers bridging probabilistic inference and neural networks. His research continues to push the boundaries of how traditional state estimation techniques can be reimagined for modern, learning-driven systems, making him a key figure in the growing area of differentiable probabilistic programming.