Deep Knowledge Transfer for Generalization Across Tasks and Domains Under Data Scarcity: On Intersections of Anomaly Detection, Few-Shot Learning, Continual Learning, Domain Generalization and Data-Free Learning
MCML Authors
Ahmed Frikha
Dr.
* Former Member
Abstract
Ahmed Frikha
Dr.
* Former Member
Abstract
This thesis addresses challenges in deep learning when key assumptions, such as abundant data or i.i.d. conditions, are violated. It introduces methods for anomaly detection with scarce data, enabling models to learn sequential tasks with minimal forgetting. For domain generalization, it proposes a feature-discovery algorithm that enhances generalization to unseen domains and a data-free approach to create robust models by synthesizing cross-domain knowledge from pre-trained models. These contributions advance deep learning for complex real-world scenarios.(Shortened).
BibTeXKey: Fri22a