sapienza university of rome · italy
Mathematics and Deep Learning
Two directions, one language.
Scope
Mathematics and deep learning shape each other. In one direction, mathematics supplies the language and guarantees that make deep learning tractable: approximation, optimization, probability, geometry, algebra, dynamical systems. In the other, deep learning has become a tool for mathematics itself: proposing conjectures, formalizing proofs, solving equations, finding structure in objects that resist classical analysis.
This workshop is for researchers working in either direction, or ideally both. The mathematical scope is intentionally open: any area of mathematics, pure or applied, plus work that turns deep learning toward mathematical discovery. Our aim is to put foundational theory and concrete practice in the same room, so that guarantees inform methods and methods raise new questions for theory.
Topics
Mathematics → Deep Learning
- Approximation theory, expressivity, and the limits of neural models
- Optimization, training dynamics, and loss landscapes
- Statistical learning theory, generalization, and uncertainty
- Geometry, topology, and algebra of representations and architectures
- Differential equations and continuous-time models (neural ODEs, diffusion, flows)
- Information-theoretic and probabilistic foundations
Deep Learning → Mathematics
- Automated and interactive theorem proving and formalization
- Conjecture generation and machine-assisted mathematical discovery
- Symbolic regression and recovery of closed-form expressions
- Neural and operator-learning solvers for PDEs and scientific computing
- Pattern and structure discovery across domains (number theory, combinatorics, representation theory, …)
On the Bridge — Both Directions
- Provable guarantees that feed back into algorithm and architecture design
- Co-design of mathematical theory and learning methods
- Benchmarks, datasets, and reproducibility for math–ML research
Important Dates
| Event | Date |
|---|---|
| Submission deadline | TBD |
| Author notification | TBD |
| Camera-ready | TBD |
| Workshop | 3–4 December 2026, Rome |