Continual Learning: Towards Image Classification From Sequential Data

Jiangpeng He, Purdue University

Abstract

Though modern deep learning based approaches have achieved remarkable progress in computer vision community such as image classification using a static image dataset, it suffers from catastrophic forgetting when learning new classes incrementally in a phase-by-phase fashion, in which only data for new classes are provided at each learning phase. In this work we focus on continual learning with the objective of learning new tasks from sequentially available data without forgetting the learned knowledge. We study this problem from three perspectives including (1) continual learning in online scenario where each data is used only once for training (2) continual learning in unsupervised scenario where no class label is provided and (3) continual learning in real world applications. Specifically, for problem (1), we proposed a variant of knowledge distillation loss together with a two-step learning technique to efficiently maintain the learned knowledge and a novel candidates selection algorithm to reduce the prediction bias towards new classes. For problem (2), we introduced a new framework for unsupervised continual learning by using pseudo labels obtained from cluster assignments and an efficient out-of-distribution detector is designed to identify whether each new data belongs to new or learned classes in unsupervised scenario. For problem (3), we proposed a novel training regime targeted on food images using balanced training batch and a more efficient exemplar selection algorithm. Besides, we further proposed an exemplar-free continual learning approach to address the memory issue and privacy concerns caused by storing part of old data as exemplars.In addition to the work related to continual learning, we study the image-based dietary assessment with the objective of determining what someone eats and how much energy is consumed during the course of a day by using food or eating scene images. Specifically, we proposed a multi-task framework for simultaneously classification and portion size estimation by future fusion and soft-parameter sharing between backbone networks. Besides, we introduce RGB-Distribution image by concatenating the RGB image with the energy distribution map as the fourth channel, which is then used for end-to-end multi-food recognition and portion size estimation.

Degree

Ph.D.

Advisors

Zhu, Purdue University.

Subject Area

Artificial intelligence|Educational technology|Pedagogy

COinS