Continual learning via inter-task synaptic mapping
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 f...
Main Authors: | , , , |
---|---|
Other Authors: | |
Format: | Journal Article |
Language: | English |
Published: |
2022
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/160691 |
_version_ | 1811689395280936960 |
---|---|
author | Mao, Fubing Weng, Weiwei Pratama, Mahardhika Yee, Edward Yapp Kien |
author2 | School of Computer Science and Engineering |
author_facet | School of Computer Science and Engineering Mao, Fubing Weng, Weiwei Pratama, Mahardhika Yee, Edward Yapp Kien |
author_sort | Mao, Fubing |
collection | NTU |
description | 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 https://github.com/ContinualAL/ISYANAKBS. |
first_indexed | 2024-10-01T05:47:25Z |
format | Journal Article |
id | ntu-10356/160691 |
institution | Nanyang Technological University |
language | English |
last_indexed | 2024-10-01T05:47:25Z |
publishDate | 2022 |
record_format | dspace |
spelling | ntu-10356/1606912022-08-01T03:24:40Z Continual learning via inter-task synaptic mapping Mao, Fubing Weng, Weiwei Pratama, Mahardhika Yee, Edward Yapp Kien School of Computer Science and Engineering Engineering::Computer science and engineering Continual Learning Lifelong Learning 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 https://github.com/ContinualAL/ISYANAKBS. National Research Foundation (NRF) This research is financially supported by National Research Foundation, Republic of Singapore under IAFPP in the AME domain (contract no.: A19C1A0018). 2022-08-01T03:24:40Z 2022-08-01T03:24:40Z 2021 Journal Article Mao, F., Weng, W., Pratama, M. & Yee, E. Y. K. (2021). Continual learning via inter-task synaptic mapping. Knowledge-Based Systems, 222, 106947-. https://dx.doi.org/10.1016/j.knosys.2021.106947 0950-7051 https://hdl.handle.net/10356/160691 10.1016/j.knosys.2021.106947 2-s2.0-85103417937 222 106947 en A19C1A0018 Knowledge-Based Systems © 2021 Elsevier B.V. All rights reserved. |
spellingShingle | Engineering::Computer science and engineering Continual Learning Lifelong Learning Mao, Fubing Weng, Weiwei Pratama, Mahardhika Yee, Edward Yapp Kien Continual learning via inter-task synaptic mapping |
title | Continual learning via inter-task synaptic mapping |
title_full | Continual learning via inter-task synaptic mapping |
title_fullStr | Continual learning via inter-task synaptic mapping |
title_full_unstemmed | Continual learning via inter-task synaptic mapping |
title_short | Continual learning via inter-task synaptic mapping |
title_sort | continual learning via inter task synaptic mapping |
topic | Engineering::Computer science and engineering Continual Learning Lifelong Learning |
url | https://hdl.handle.net/10356/160691 |
work_keys_str_mv | AT maofubing continuallearningviaintertasksynapticmapping AT wengweiwei continuallearningviaintertasksynapticmapping AT pratamamahardhika continuallearningviaintertasksynapticmapping AT yeeedwardyappkien continuallearningviaintertasksynapticmapping |