Kursinformation

20-00-1118-iv Human and Identity centric Machine Learning

• Instructors: Naser Damer
• Event type: Integrated Course
• Org-unit: Dept. 20 - Computer Science
• Crediting for: Hours per week: 4
• Language of instruction: English
• Min. | Max. participants: - | -

Course Contents

• Background and concepts of human-centric Machine Learning: the goal of identity and human-centric machine learning. The differences between identity learning and other mainstream classification.
• Representation extraction for subject-related data: feature extraction methodology for identity related applications. Hand crafted and Deeply learned features background and basics.
• Deep learning strategies for identity representations: learning identities representations with deep learning. Learning strategies and learning losses. Network architectures and identityspecific components.
• Knowledge transfer and distillation: transfer learning and identity-representation. Knowledge distillation concepts and applications.
• Efficient machine learning: the relation between resource limitations, Green-AI, and deep learning. Methods to build efficient machine learning solutions.
• Synthetic identity: the need of synthetic identity. Synthetic identity as adversarial. Generating synthetic identity-controlled data under different restrictions.
• Machine learning biases: analyses of demographic fairness and the roots of the fairness issues. ML-based mitigation of demographic biases
• Learning privacy: analyzing unintentionally learned information. Learning strategies to the targeted suppression of information at different representation levels.
• Data utility: understanding the effect of data utility in the training process. Understanding sample utility in operation. ML concepts and strategies of estimating sample utilities.
• Sample-level attacks: overview on adversarial, sample manipulation, other attacks on human-centric ML. Deep learning concepts, network blocks, and loss strategies, to detect and mitigate sample-level attacks.
• Explainability: overview on the need for explainability in different decision-making processes. Different strategies to provide explainability for decision made in different operations discussed in the previous lectures.
• Ethics in identity-centric ML: overview on ethics in AI and AI regulation. AI ethics for human data processing and storage.