CVPR 2022 Tutorial on

Deep AUC Maximization: From Algorithms to Practice

8:45am - 12:15pm, June 20, 2022


Tutorial Lecturers




Overview

Areas under the curves (AUC), including area under ROC curve (AUROC) and area under Precision-Recall curve (AUPRC) are widely used performance measures in many important computer vision and AI tasks and have extraordinarily varied applications in information retrieval, object detection, medical diagnosis, etc. AUROC is traditionally used as a measure for binary classification with imbalanced data. It has been used in many vision tasks, such as face verification, medical image classification. AUPRC or its variant called average precision (AP) is another popular measure for image/text retrieval, objective detection. Recently, there is a surge of research activities on developing efficient stochastic optimization algorithms and methods for deep AUC maximization, which directly minimize a surrogate loss of AUC for learning a deep neural network. The main objectives of this tutorial are three-fold:

Schedule

• Introduction (motivations, definitions, history, structure) (8:45 - 9:00am, Yang)

• Old-school Methods (Batch and Online Methods) (9:00 - 9:15am, Ying)

• AUC Maximization in the Era of Big Data (9:15 - 9:45am, Ying)

• AUC Maximization in the Era of AI (9:45 - 10:15am, Liu)

• Break (10:15 - 10:30am)

• Partial AUC Maximization: Old and New (10:30 - 11:00am, Narasimhan)

• AP Maximization (11:00 - 11:30am, Yang)

• Applications & Library (11:30 - 11:50am, Yang)

• Future Work (11:50 - 12:15pm, Yang)


Resources