Adversarial Feature Alignment: Avoid Catastrophic Forgetting in Incremental Task Lifelong Learning

DOI10.1162/NECO_A_01232MaRDI QIDQ5214401zbMATH OpenOpenAlexDBLPWikidataFDO

Authors Xin Yao, Tianchi Huang, Chenglei Wu, Rui-Xiao Zhang, Li-Feng Sun

Publication date 7 February 2020

Published in Neural Computation (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1910.10986

zbMATH Keywords

catastrophic forgettingincremental multitask image classificationincremental task lifelong learningneural network model

Mathematics Subject Classification ID

Memory and learning in psychology (91E40) Neural networks for/in biological studies, artificial life and related topics (92B20)

Abstract: Human beings are able to master a variety of knowledge and skills with ongoing learning. By contrast, dramatic performance degradation is observed when new tasks are added to an existing neural network model. This phenomenon, termed as emph{Catastrophic Forgetting}, is one of the major roadblocks that prevent deep neural networks from achieving human-level artificial intelligence. Several research efforts, e.g. emph{Lifelong} or emph{Continual} learning algorithms, have been proposed to tackle this problem. However, they either suffer from an accumulating drop in performance as the task sequence grows longer, or require to store an excessive amount of model parameters for historical memory, or cannot obtain competitive performance on the new tasks. In this paper, we focus on the incremental multi-task image classification scenario. Inspired by the learning process of human students, where they usually decompose complex tasks into easier goals, we propose an adversarial feature alignment method to avoid catastrophic forgetting. In our design, both the low-level visual features and high-level semantic features serve as soft targets and guide the training process in multiple stages, which provide sufficient supervised information of the old tasks and help to reduce forgetting. Due to the knowledge distillation and regularization phenomenons, the proposed method gains even better performance than finetuning on the new tasks, which makes it stand out from other methods. Extensive experiments in several typical lifelong learning scenarios demonstrate that our method outperforms the state-of-the-art methods in both accuracies on new tasks and performance preservation on old tasks.

Recommendations

• Toward training recurrent neural networks for lifelong learning
• Lifelong learning in costly feature spaces
• Lifelong learning in costly feature spaces
• Hierarchically structured task-agnostic continual learning
• Progressive learning: a deep learning framework for continual learning

Cites work

• Domain-adversarial training of neural networks
• Overcoming catastrophic forgetting in neural networks

Cited in

(8)

• Progressive learning: a deep learning framework for continual learning
• scientific article; zbMATH DE number 2033143 (Why is no real title available?)
• Overcoming catastrophic forgetting in neural networks
• Toward training recurrent neural networks for lifelong learning
• Dynamic Consolidation for Continual Learning
• A comprehensive study of class incremental learning algorithms for visual tasks
• Hierarchically structured task-agnostic continual learning
• Bio-inspired, task-free continual learning through activity regularization

Describes a project that uses

Uses Software

• ImageNet
• CORe50
• iCaRL
• AlexNet
• CIFAR
• Caltech-UCSD Birds

This page was built for publication: Adversarial Feature Alignment: Avoid Catastrophic Forgetting in Incremental Task Lifelong Learning

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5214401)