Retrospective Feature Estimation for Continual Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Nguyen, Nghia D., Nguyen, Hieu Trung, Li, Ang, Pham, Hoang, Nguyen, Viet Anh, Doan, Khoa D.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912864465321984
author Nguyen, Nghia D.
Nguyen, Hieu Trung
Li, Ang
Pham, Hoang
Nguyen, Viet Anh
Doan, Khoa D.
author_facet Nguyen, Nghia D.
Nguyen, Hieu Trung
Li, Ang
Pham, Hoang
Nguyen, Viet Anh
Doan, Khoa D.
contents The intrinsic capability to continuously learn a changing data stream is a desideratum of deep neural networks (DNNs). However, current DNNs suffer from catastrophic forgetting, which interferes with remembering past knowledge. To mitigate this issue, existing Continual Learning (CL) approaches often retain exemplars for replay, regularize learning, or allocate dedicated capacity for new tasks. This paper investigates an unexplored direction for CL called Retrospective Feature Estimation (RFE). RFE learns to reverse feature changes by aligning the features from the current trained DNN backward to the feature space of the old task, where performing predictions is easier. This retrospective process utilizes a chain of small feature mapping networks called retrospector modules. Empirical experiments on several CL benchmarks, including CIFAR10, CIFAR100, and Tiny ImageNet, demonstrate the effectiveness and potential of this novel CL direction compared to existing representative CL methods, motivating further research into retrospective mechanisms as a principled alternative for mitigating catastrophic forgetting in CL. Code is available at: https://github.com/mail-research/retrospective-feature-estimation.
format Preprint
id arxiv_https___arxiv_org_abs_2406_17381
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Retrospective Feature Estimation for Continual Learning
Nguyen, Nghia D.
Nguyen, Hieu Trung
Li, Ang
Pham, Hoang
Nguyen, Viet Anh
Doan, Khoa D.
Machine Learning
Computer Vision and Pattern Recognition
The intrinsic capability to continuously learn a changing data stream is a desideratum of deep neural networks (DNNs). However, current DNNs suffer from catastrophic forgetting, which interferes with remembering past knowledge. To mitigate this issue, existing Continual Learning (CL) approaches often retain exemplars for replay, regularize learning, or allocate dedicated capacity for new tasks. This paper investigates an unexplored direction for CL called Retrospective Feature Estimation (RFE). RFE learns to reverse feature changes by aligning the features from the current trained DNN backward to the feature space of the old task, where performing predictions is easier. This retrospective process utilizes a chain of small feature mapping networks called retrospector modules. Empirical experiments on several CL benchmarks, including CIFAR10, CIFAR100, and Tiny ImageNet, demonstrate the effectiveness and potential of this novel CL direction compared to existing representative CL methods, motivating further research into retrospective mechanisms as a principled alternative for mitigating catastrophic forgetting in CL. Code is available at: https://github.com/mail-research/retrospective-feature-estimation.
title Retrospective Feature Estimation for Continual Learning
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2406.17381