Conference on the Mathematical Theory of Deep Neural Networks 2025

Important Dates

• Submissions open:​ July 1, 2025
• Notifications of decision sent:​ Oct 1, 2025​
• Submissions close:​ Aug 31, 2025
• Registration opens: Oct 1, 2025
• Registration deadline: Nov 5, 2025
• Conference: Nov 6-7, 2025

General info

Recent advances in deep neural networks (DNNs), combined with open, easily-accessible implementations, have made DNNs a powerful, versatile method used widely in both machine learning and neuroscience. These advances in practical results, however, have far outpaced a formal understanding of these networks and their training. The dearth of rigorous analysis for these techniques limits their usefulness in addressing scientific questions and, more broadly, hinders systematic design of the next generation of networks. Recently, long-past-due theoretical results have begun to emerge from researchers in a number of fields. The purpose of this conference is to give visibility to these results, and those that will follow in their wake, to shed light on the properties of large, adaptive, distributed learning architectures, and to revolutionize our understanding of these systems.

Topics and Submission

Investigators interested in having their abstracts considered for presentation should submit their abstracts no later than August 31, 2025.

DeepMath is a highly interdisciplinary conference focused on understanding fundamental theory driving the success of Deep Learning. A principal goal of this conference is to bring together theoreticians working on deep learning from various disciplines and perspectives. We, therefore, encourage submissions from researchers from diverse disciplines including but not limited to

• ​​Statistics
• Physics
• Computer science
• Neuroscience
• Mathematics
• Psychology
• Engineering

Topics may address any area of deep learning research such as:

• Expressivity
• Generalization
• Optimization
• Representations
• Computation
• Network architectures​
• Recurrent networks

To complement the many conferences with applications and theory the focus for DeepMath will be exclusively on the theoretical and mechanistic understanding of the underlying properties of neural networks.

Abstracts will not be made public (i.e., no official proceedings), and will be doubly-blind reviewed and selected for quality. All poster submissions should be properly anonymized in order to allow for blind refereeing. Submissions should be no more than 1 page although a second page may be used for references. Authors should submit a pdf file prepared using the Latex style file available here and should adopt all formatting, subject headings, font sizes, etc. defined therein. Submissions that fail to meet the format requirements will not be reviewed. The first author listed on the abstract is considered to be the presenting author. Each presenting author may submit only one abstract.

Source: https://deepmath-conference.com/