Please use this identifier to cite or link to this item:
Title: Continual learning via inter-task synaptic mapping
Authors: Mao, Fubing
Weng, Weiwei
Pratama, Mahardhika
Yee, Edward Yapp Kien
Keywords: Engineering::Computer science and engineering
Issue Date: 2021
Source: Mao, F., Weng, W., Pratama, M. & Yee, E. Y. K. (2021). Continual learning via inter-task synaptic mapping. Knowledge-Based Systems, 222, 106947-.
Project: A19C1A0018
Journal: Knowledge-Based Systems 
Abstract: Learning from streaming tasks leads a model to catastrophically erase unique experiences it absorbs from previous episodes. While regularization techniques such as LWF, SI, EWC have proven themselves as an effective avenue to overcome this issue by constraining important parameters of old tasks from changing when accepting new concepts, these approaches do not exploit common information of each task which can be shared to existing neurons. As a result, they do not scale well to large-scale problems since the parameter importance variables quickly explode. An Inter-Task Synaptic Mapping (ISYANA) is proposed here to underpin knowledge retention for continual learning. ISYANA combines task-to-neuron relationship as well as concept-to-concept relationship such that it prevents a neuron to embrace distinct concepts while merely accepting relevant concept. Numerical study in the benchmark continual learning problems has been carried out followed by comparison against prominent continual learning algorithms. ISYANA exhibits competitive performance compared to state of the arts. Codes of ISYANA is made available in
ISSN: 0950-7051
DOI: 10.1016/j.knosys.2021.106947
Schools: School of Computer Science and Engineering 
Rights: © 2021 Elsevier B.V. All rights reserved.
Fulltext Permission: none
Fulltext Availability: No Fulltext
Appears in Collections:SCSE Journal Articles

Citations 20

Updated on May 25, 2023

Web of ScienceTM
Citations 20

Updated on May 27, 2023

Page view(s)

Updated on Jun 1, 2023

Google ScholarTM




Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.