Organisation
Project Advanced Visual Deep Learning, summer term 2025
Description
The project focuses on a deep exploration of state-of-the-art visual deep learning architectures, including transformer-based models, diffusion models, and multi-modal networks. Students will work on individual or small-group projects where they will not only train and fine-tune models but also actively modify network structures, optimize the training process, and implement novel variations of existing architectures. Using PyTorch and the VisCom GPU cluster as well as the BwUniCluster, they will conduct large-scale experiments and systematically evaluate their models. The course emphasizes a structured approach, incorporating software engineering principles, systematic experimentation, and model interpretability. Throughout the module, students will develop the ability to critically assess and improve deep learning architectures, gaining valuable experience for both research and industry applications.
Modalities
IIn this project, students will develop an in-depth understanding of advanced deep learning architectures for visual data. Beyond training models and handling datasets, they will critically analyze, modify, and optimize deep neural network architectures to improve performance, efficiency, and interpretability. They will explore key concepts such as vision transformers, generative models, self-supervised learning, and efficient training techniques. By the end of the module, students will have gained the skills to not only apply existing deep learning methods but also to adapt and extend architectures to tackle complex real-world challenges in a research-driven manner.
