Module MA4320-KP05

Optimisation methods for machine learning (OptvML)


Duration

1 Semester

Turnus of offer

irregularly

Credit points

5

Course of studies, specific fields and terms:

  • Bachelor CLS 2016, optional subject, mathematics
  • Master CLS 2016, optional subject, mathematics
  • Bachelor CLS 2023, optional subject, mathematics
  • Master CLS 2023, optional subject, mathematics

Classes and lectures:

  • MA4320-Ü: Optimisation methods for machine learning (exercise, 1 SWS)
  • MA4320-V: Optimisation methods for machine learning (lecture, 2 SWS)

Workload:

  • 45 hours in-classroom work
  • 85 hours private studies and exercises
  • 20 hours exam preparation

Contents of teaching:

  • Fundamentals of current machine learning methods and their applications in Artificial Intelligence
  • First-order stochastic optimization methods, moment-based optimization methods
  • Continuous-time modeling, Langevin dynamics
  • Adaptive and moment-based optimization methods
  • Higher-order optimization methods
  • Deep learning from the perspective of optimal control
  • Optimization on manifolds
  • Information-geometric approaches

Qualification-goals/Competencies:

  • Students understand the fundamentals of current machine learning methods
  • Students have an overview of optimization methods for high-dimensional problems and know the mathematical tools for their analysis
  • Students can view optimization methods from different mathematical perspectives
  • Students have a basic understanding of differential and information geometry and their use in numerical optimization
  • Interdisciplinary aspects:
  • They can translate theoretical concepts into practical solutions.
  • They have implementation experience.
  • They are able to abstract practical problems.

Grading through:

  • Written or oral exam as announced by the examiner

Responsible for this module:

Literature:

  • Bottou, Curtis, Nocedal : Optimization Methods for Large-Scale Machine Learning SIAM
  • Absil, Mahony, Sepulchre : Optimization Algorithms on Matrix Manifolds
  • Amari : Information Geometry and Its Applications

Language:

  • German and English skills required

Notes:

Admission requirements for taking the module:
- None (the competences of the modules mentioned under ‘Requires’ are needed for this module, but are not a formal prerequisite)

Admission requirements for participation in module examination(s):
- Successful completion of exercises and their presentation in accordance with the specifications at the beginning of the semester

Module Exam(s):
- MA4320-L1: Optimisation methods for machine learning, written exam, 90min, or oral exam, 30min, according to the lecturer, 100% of the module grade

Last Updated:

09.02.2026