Convergence Analysis
Convergence analysis in machine learning focuses on rigorously proving the stability and efficiency of algorithms used to train complex models. Current research emphasizes developing and analyzing tuning-free algorithms for bilevel optimization and addressing convergence challenges in various architectures, including variational autoencoders, physics-informed neural networks, and deep sparse coding models, often leveraging techniques like neural tangent kernels and variable splitting. These advancements enhance the theoretical understanding of deep learning, leading to more robust and efficient training methods with improved performance guarantees for diverse applications.
Papers
Conditional simulation via entropic optimal transport: Toward non-parametric estimation of conditional Brenier maps
Ricardo Baptista, Aram-Alexandre Pooladian, Michael Brennan, Youssef Marzouk, Jonathan Niles-Weed
Sketched Adaptive Federated Deep Learning: A Sharp Convergence Analysis
Zhijie Chen, Qiaobo Li, Arindam Banerjee