Lecture Course: Introduction to Deep Learning using PyTorch, Summer Term 2026

Fundamental Understanding: By the end of this course, students will have a comprehensive understanding of the fundamental concepts and theories that underlie deep learning, including neural network architectures, activation and loss functions, and backpropagation. They will understand how these components form the computational basis for perception, representation learning, and intelligent behavior.

PyTorch Proficiency: Students will be able to effectively use PyTorch to build, train, and evaluate deep learning models. They will acquire the skills to design and implement custom neural architectures, from perception models to networks that learn internal world representations, and to interpret their behavior through experimentation.

Model Development and Optimization: Upon completing the course, students will be able to use PyTorch to investigate and prototype models that both perceive and act, for example: vision systems, generative models, or learning agents. They will learn to refine model architectures and hyperparameters to improve robustness, generalization, and adaptive behavior.

In recent years, PyTorch has become one of the most widely used frameworks for developing and experimenting with deep learning models. This course provides a practical and conceptual introduction to deep learning with PyTorch, covering both perceptual and agentic aspects of modern deep learning.

The first part of the course introduces the mathematical and computational foundations of neural networks, including feed-forward, convolutional, recurrent, and transformer architectures. Students learn how to implement, train, and analyze networks using PyTorch, and how these models can be applied to real-world tasks in computer vision, language, and in building systems that perceive, reason, and act as intelligent agents.

In the project phase, students choose between two complementary tracks:                                   - a vision track, focusing on the application of CNNs and Transformers on generative and perceptual tasks,                                                                                                                                     - an agent track, exploring world models, intrinsic motivation, and goal-conditioned learning, where agents learn to model and understand their environment through interaction.

By the end of the course, participants will have a solid theoretical and practical foundation in deep learning and will be equipped to develop models that perceive, learn, and adapt across diverse domains using PyTorch.