Workshop 04

Resource Person:

Dr. Fahad Sohrab

Overview:

This workshop introduces one-class classification (OCC), a learning paradigm for scenarios where training data are available only from a single target class, while other classes are scarce, unknown, or difficult to obtain. It discusses the limitations of traditional OCC methods, particularly in handling high-dimensional and complex data, and presents subspace learning-based OCC as an effective and principled solution. The workshop covers core theoretical foundations, optimization frameworks, and recent advances, including joint learning of subspaces and data descriptions, graph-based formulations, and multi-view extensions. Participants will gain practical insights through case studies and demonstrations, illustrating how these methods can be applied to anomaly detection and data-scarce learning problems across domains.

Objectives:

• Understand the fundamentals and challenges of one-class classification (OCC).
• Identify limitations of conventional OCC methods in high-dimensional and complex data settings.
• Learn the principles of subspace learning-based OCC.
• Explore joint optimization of subspace representations and data descriptions.
• Understand graph-based and multi-view extensions of OCC methods.
• Apply OCC techniques to anomaly detection and data-scarce scenarios.
• Gain practical insights through examples and demonstrations.
• Develop the ability to select and adapt subspace learning-based OCC methods for different applications.