Deep learning has achieved remarkable success in numerous practical domains, yet the theoretical understanding of its core mechanisms remains fragmented and inadequate. Current theoretical frameworks often treat deep neural networks (DNNs) as simple large-scale parameterized models, failing to effectively integrate approximation error, optimization error, and the generalization gap, and thus cannot fully explain the complex dynamics of high-dimensional and overparameterized DNNs in practical applications. This disconnect between theory and practice has become a key bottleneck restricting the development of more reliable, interpretable, and efficient deep learning systems. Focused on the theme of “theoretical analysis of deep learning,” this half-day workshop aims to build a high-level academic exchange platform for researchers in theoretical AI, machine learning, and applied neural networks. The workshop will focus on cutting-edge topics such as integrated error analysis in deep learning, theoretical explorations of sparse feature learning, convergence guarantees of training algorithms, and generalization dynamics of DNNs—core directions closely aligned with the latest research frontiers. By bringing together both established experts and early-career researchers, the workshop will showcase innovative theoretical frameworks, rigorous analytical methods, and their potential practical implications. Through a combination of invited talks, panel discussions, and interactive brainstorming sessions, the workshop will promote in-depth discussion of unresolved theoretical issues, encourage cross-fertilization of ideas between approximation theory, learning theory, and optimization theory in the context of deep learning, and help identify new research directions that bridge the gap between theory and practice. The workshop is designed to have a broader scope than a typical special session, covering both foundational theoretical developments and practice-oriented theoretical insights. It will provide a valuable opportunity for participants to share research results, discuss technical challenges, and establish academic collaborations, thereby advancing theoretical research in deep learning and providing solid theoretical support for the innovation and application of deep learning technologies.
Workshop Organizers
- Shuo Huang
- Han Feng
- Jun Fan
- Yunwei Lei