Modul
Numerical Analysis of Neural Networks [M-MATH-106695]
Credits
6Recurrence
UnregelmäßigDuration
1 SemesterLanguage
German/EnglishLevel
4Version
1Responsible
Organisation
- KIT-Fakultät für Mathematik
Part of
Bricks
Identifier | Name | LP |
---|---|---|
T-MATH-113470 | Numerical Analysis of Neural Networks | 6 |
Competence Certificate
The module will be completed by an oral exam (about 30 min).
Competence Goal
The goal of the lecture is to provide a mathematical foundation of neural networks from the perspective of numerical analysis. Students know basic definitions and terminology as well as classical approximation results for neural networks. They are familiar with numerical methods for the efficient training of neural networks and can analyze them. Moreover, students can apply the concepts to popular applications in the context of partial differential equations (such as physics-informed neural networks).
Prerequisites
none
Content
- Neural networks
- Approximation results
- Connections to finite element methods
- Numerical methods for the efficient learning
- Physics-informed neural networks
Recommendation
A solid background in numerical mathematics is strongly recommended. Basic knowledge of functional analysis and finite element methods is helpful, but not required.
Workload
Total workload: 180 hours
Attendance: 60 h
- lectures, problem classes and examination
Self studies: 120 h
- follow-up and deepening of the course content,
- work on problem sheets,
- literature study and internet research on the course content,
- preparation for the module examination