|
|
|
| Module code: KIM-GKR |
|
2V+2P (4 hours per week) |
|
6 |
| Semester: 2 or 3 |
| Mandatory course: no |
Language of instruction:
German |
Assessment:
Oral examination (50%) und project work (50%)
[updated 29.10.2026]
|
E2944 Electrical Engineering and Information Technology, Master, regulation 01.04.2019
, optional course
FTM-GKR Automotive Engineering, Master, regulation 01.04.2023
, optional course
KIM-GKR Computer Science and Communication Systems, Master, regulation 01.10.2019
, optional course
MAM.GKR Engineering and Management, Master, regulation 01.10.2019
, optional course
MP3206.GKR Medical Physics, Master, regulation 01.10.2025
, optional course
PIM-GKR Applied Informatics, Master, regulation 01.10.2026
, semester 2, optional course
TIM-GKR Technical Computer Science, Master, regulation 01.10.2026
, optional course
Suitable for exchange students (learning agreement)
|
60 class hours (= 45 clock hours) over a 15-week period.
The total student study time is 180 hours (equivalent to 6 ECTS credits).
There are therefore 135 hours available for class preparation and follow-up work and exam preparation.
|
Recommended prerequisites (modules):
None.
|
Recommended as prerequisite for:
|
Module coordinator:
Prof. Dr.-Ing. Ahmad Osman |
Lecturer: Prof. Dr.-Ing. Ahmad Osman
[updated 16.07.2026]
|
Learning outcomes:
This module covers advanced methods in artificial intelligence, with a focus on generative AI and risk-aware intelligent systems.
After successfully completing this module, students will:
• be able to analyze and implement advanced neural architectures in practice,
• understand and be able to use generative models (for example: GANs, diffusion models, transformer architectures, large language models (LLMs)),
• be able to distinguish between and quantify aleatory and epistemic uncertainties in AI systems,
• be able to perform model calibration and confidence estimation,
• be able to evaluate intelligent systems under realistic operational conditions (domain shift, data scarcity, regulatory requirements),
• be able to integrate risk-based decision-making mechanisms and human-in-the-loop concepts,
• be able to model industrial use cases in a structured way and develop AI solutions ready for deployment.
The module combines modern generative AI with uncertainty modeling, decision theory, and industrial implementation.
[updated 29.10.2026]
|
Module content:
Part I - In-Depth Study of Advanced ML Methods
• Representational learning
• Ensemble methods and model calibration
• Fundamentals of probabilistic modelling
Part II - Generative models
• Generative Adversarial Networks (GANs)
• Diffusion models
• Transformer architectures
• Large Language models (LLMs)
• Multimodal generative systems
Part III - Uncertainty and Risk in AI Systems
• Aleatoric vs. epistemic uncertainty
• Confidence estimation and calibration
• Set-valued / Imprecise models
• Decision making under uncertainty
Part IV - Deployment and Governance
• Domain shift and robustness
• Evaluation metrics for high-risk applications
• Human-in-the-loop systems
• Fundamentals of regulatory frameworks (e.g., the EU AI Act)
• Project-based work with practical applications
[updated 29.10.2026]
|
Teaching methods/Media:
Lectures with interactive discussions, a structured intensive phase (block format), practical implementation of modern AI methods, project-based group work with academic reports, presentations, and current international scholarly literature.
[updated 29.10.2026]
|
Recommended or required reading:
Murphy, K. P.: Probabilistic Machine Learning - MIT Press, 2022
Bishop, C.: Deep Learning - Springer, 2023
Transformers / LLM foundations: Vaswani et al., “Attention Is All You Need”
Diffusion models (core GenAI pillar): Ho et al., “Denoising Diffusion Probabilistic Models (DDPM)”
GANs (still important historically + conceptually): Goodfellow et al., “Generative Adversarial Nets”
Uncertainty + risk-aware AI (matches your Epistemic/Imprecise positioning):
• Kendall & Gal, “What Uncertainties Do We Need…?” (aleatoric vs epistemic)
• Lakshminarayanan et al., “Deep Ensembles”
• Guo et al., “On Calibration of Modern Neural Networks”
Goodfellow, I., Bengio, Y., Courville, A.: Deep Learning - MIT Press, 2016
Shalev-Shwartz, S., Ben-David, S.: Understanding Machine Learning - Cambridge University Press, 2014
Ausgewählte aktuelle wissenschaftliche Veröffentlichungen zu Generative AI, Unsicherheitsquantifizierung und industrieller KI (werden im Kurs bereitgestellt).
Independent research is also part of the project work.
Transformer (Attention Is All You Need) PDF:
https://arxiv.org/pdf/1706.03762
DDPM (Diffusion) PDF:
https://arxiv.org/pdf/2006.11239
GANs (Generative Adversarial Nets) PDF:
https://arxiv.org/pdf/1406.2661
Deep Ensembles PDF:
https://arxiv.org/pdf/1612.01474
Calibration (Guo et al.) PDF:
https://arxiv.org/pdf/1706.04599
Kendall & Gal uncertainties PDF:
https://arxiv.org/pdf/1703.04977
[updated 29.10.2026]
|
