The Human-Centered Machine Learning (HCML) module conveys the theoretical and practical foundations for developing machine learning systems in which humans occupy a central role throughout the entire ML lifecycle. The focus lies on the training, adaptation, evaluation, and reflection of ML models, with explicit consideration of human needs, capabilities, limitations, and values.

The course covers fundamental and advanced concepts of machine learning, with an emphasis on neural networks. Classical machine learning methods (e.g., Support Vector Machines, Random Forests) are utilized as baselines. Building upon these, neural networks are treated systematically, ranging from data preprocessing, feature engineering, and representation learning to neural network architecture, backpropagation, optimization methods, and hyperparameter selection.

Topics covered include, among others:

• Fundamentals of Machine Learning, specifically supervised and unsupervised learning
• Data Preprocessing, feature engineering, and representation learning
• Architecture and Functionality of neural networks
• Training, Validation, and Testing Strategies
• Overfitting, Underfitting, and regularization
• Hyperparameter Tuning and model selection
• Utilization and Adaptation of pre-trained models
• Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM)
• Generative Adversarial Networks (GANs)
• Evaluation of ML Models and appropriate metrics
• Interpretability and Critical Analysis of model decisions in human-centered application contexts

After successfully completing the module, students will be able to:

• To identify, explain, and contextualize central concepts, methods, and terminology of machine learning—specifically neural networks—within the framework of human-centered applications.
• To systematically collect, prepare, and analyze data for ML systems, and to critically reflect on the impact of data quality, bias, and representativeness on model behavior.
• To train, configure, and optimize ML models, including the selection of appropriate architectures, training strategies, and hyperparameters.
• To compare different model classes and learning paradigms and evaluate their suitability for specific human-centered application scenarios.
• To evaluate ML models using appropriate metrics and critically interpret results with regard to generalizability, robustness, and traceability.
• To analyze model decisions and limitations in an understandable manner and discuss their consequences for users as well as for deployment in real-world contexts.
• To implement ML-based systems as prototypes, develop them iteratively, and present as well as reflect on the results in a structured way.
• To critically reflect on one's own design, modeling, and evaluation decisions in order to derive principles for a scientifically sound and responsible design and use of machine learning systems.

The accompanying exercises deepen the course content through practical implementations in Python and Jupyter Notebooks. Students will train, evaluate, and compare their own models using concrete examples and develop their own project over the course of the semester. Interim results are presented, reflected upon, and iteratively improved. Consequently, the module combines a solid methodological foundation in machine learning with a clear human-centered perspective on the design and evaluation of learning systems.

Module examination: 90 minutes written examination or oral examination