In this lecture we took a closer look at selected topics in deep learning with a focus on computer vision, covering theoretical and practical aspects. We discussed design principles and properties of selected neural networks starting with classifiers and then extending to selected applications like object recognition, image registration and also discuss the application of computer vision based technologies to non-standard visual domains like scanned documents and spectrograms. In the second part of the course we turn to the more theoretical works of the field, addressing currently open questions. We exemplarily discussed topics such as deep generative models, adversarial examples, introspection, and explainability. Finally we provided practical guidelines for developing deep learning solutions based on the current state of research as well industry experience.

Transfer Learning

Transfer learning is a machine learning technique where knowledge gained from training a model on one task is leveraged to improve performance on a different but related task. Instead of training a model from scratch on a new task, transfer learning allows us to use pre-trained models that have already learned useful features from large-scale datasets.You can find the slides here.

Similarity learning

Similarity learning, also known as metric learning, is a subfield of machine learning that focuses on learning a function that measures the similarity or dissimilarity between data samples. The goal of similarity learning is to transform the data into a new representation where similar samples are mapped closer together while dissimilar samples are mapped farther apart.You can see the slides here.