Summer School Course
Date: September 20 – October 2, 2026
Location: Sarntal Valley, South Tyrol, Italy
Instructors: Prof. Dr. Felix Dietrich (TUM) and Prof. Dr. Christian Riess (FAU)
Course Description
In this course we will explore techniques addressing either security or explainibility of machine learning models.
The course is organized into six parts, we will first discuss basic and advanced machine learning techniques before diving deeper into different specializations of secure and explainable ML. Each participant will give a 30 minute presentation on a selected topic from one of these six blocks, and contribute to a practice project.
Course Topics
- Basic Machine Learning
This block covers the basic algorithms of modern machine learning. Specific topics include the core components of neural networks and how to train them from data. In addition to the technical foundations we will discuss practical issues such as sustainability, societal impact and policy/regulation. This block contains topics that are well approachable without any prior background in machine learning.
- Advanced Machine Learning
This block covers selected advanced neural architectures and training methods. The focus will be on the inner workings of what is colloquially mentioned as “AI”. Individual topics cover the architectures and algorithms that enable Large Language Models, as well as what heat transfer has to do with image generation with diffusion models. Additionally, we will discuss how specialised models and software-hardware co-design can enable energy-efficient machine learning.
- Scientific Machine Learning
This block covers how to apply machine learning to problems from science & engineering, with a focus on algorithms. Individual topics include how knowledge from physics can be embedded into models and algorithms to solve problems in a structured and data-efficient manner. We also discuss how underlying physical laws can be discovered from observations using observations of the process. Towards efficient machine learning methods, we discuss how randomness can enable extremely fast training of neural networks.
- Applications of SciML
This block covers how to apply machine learning to problems from science & engineering, with a focus on case-studies and domain-specific problems. Specific topics cover how machine learning accelerates costly simulations and solves otherwise intractable inverse problems. For this, we will look at applications from various domains ranging from fluid dynamics and medical imaging to robotics.
- Machine Learning Security
This block covers selected aspects on the security and privacy of machine learning models. Specific topics will be attacks during training and test time and specific attacker goals such as compromising either the integrity of the computation or the privacy of the data. We will also look at defenses for machine learning systems and discuss their viability.
- Attribution of Generated Content
This block covers different methods for identifying artificially generated content in text and images. Hereby, we will be looking at passive and active techniques. Passive detection mechanisms forensically examine AI-generated content and identify features that help in distinguishing between photographs/human-written text and generated images/machine-written text. Active mechanisms embed watermarks as an imperceptible signal in generated text or images.