Jiang Hu

alt text 

Postdoc at UC Berkeley.
867 Evans Hall,
Berkeley, CA 94720
Email: hujiangopt@gmail.com, jianghu@berkeley.edu

About me

I am a postdoc at the Department of Mathematics, University of California, Berkeley. My current research focuses on manifold optimization, nonsmooth optimization, decentralized optimization and federated learning, and their applications in machine learning and signal processing.

News

  • [3/2025] One paper titled ‘‘On the local convergence of the semismooth Newton method for composite optimization’’ is accepted by Journal of Scientific Computing. [link]

  • [2/2025] One paper titled ‘‘Achieving Local Consensus over Compact Submanifolds’’ is accepted by IEEE Transactions on Automatic Control. [link]

  • [1/2025] One paper titled ‘‘An Augmented Lagrangian Primal-Dual Semismooth Newton Method for Multi-block Composite Optimization’’ is accepted by Journal of Scientific Computing. [link]

  • [12/2024] One paper titled ‘‘Decentralized projected Riemannian stochastic recursive momentum method for nonconvex optimization’’ is accepted by AAAI 2025. [link]

  • [9/2024] One paper titled ‘‘Nonconvex Federated Learning on Compact Smooth Submanifolds With Heterogeneous Data’’ is accepted by NeurIPS 2024. [link]

  • [7/2024] Our manuscript titled ‘‘Improving the communication in decentralized manifold optimization through single-step consensus and compression’’ is on arxiv. [arxiv]

  • [5/2024] One paper titled ‘‘Convergence analysis of an adaptively regularized natural gradient method’’ is accepted by IEEE Transaction on Signal Processing. [link]

  • [4/2024] Honored to receive the Best Paper Award at ICASSP 2024, 1 out of 2826 accepted papers. [link] [page]

  • [2/2024] One paper titled ‘‘A projected semismooth Newton method for a class of nonconvex composite programs with strong prox-regularity’’ is accepted by Journal of Machine Learning Research. [link]

  • [1/2024] One paper titled ‘‘Riemannian Natural Gradient Methods’’ is published at SIAM Journal on Scientific Computing. [link]