Raphaël BAENA Thesis defense

The thesis focuses on the concept of generalization, particularly in the context of supervised machine learning classification. This approach involves learning to solve a task (classification) based on labeled training data. Generalization is defined as the ability to make
accurate predictions on unseen data during training. Traditionally, generalization is limited to data that belongs to the same domain as the training task. However, recent literature highlights the capability of deep learning architectures to generalize beyond their training task. Thus, a model trained on a specific task can be partially reused for other tasks. The thesis explores various possible extensions of generalization, including learning on a set of classes and the ability to generalize to a larger set of classes, learning on coarse labels to predict more complex labels, learning on artificially complex tasks to improve generalization, and learning invariant operators for a specific task.