Publicacions CVC
Home
|
Show All
|
Simple Search
|
Advanced Search
|
Add Record
|
Import
You must login to submit this form!
Login
Quick Search:
Field:
main fields
author
title
publication
keywords
abstract
created_date
call_number
contains:
...
Edit the following record:
Author
...
is Editor
Title
...
Type
Journal Article
Abstract
Book Chapter
Book Whole
Conference Article
Conference Volume
Journal
Magazine Article
Manual
Manuscript
Map
Miscellaneous
Newspaper Article
Patent
Report
Software
Year
...
Publication
...
Abbreviated Journal
...
Volume
...
Issue
...
Pages
...
Keywords
...
Abstract
In this paper, we investigate the continual learning of Vision Transformers (ViT) for the challenging exemplar-free scenario, with special focus on how to efficiently distill the knowledge of its crucial self-attention mechanism (SAM). Our work takes an initial step towards a surgical investigation of SAM for designing coherent continual learning methods in ViTs. We first carry out an evaluation of established continual learning regularization techniques. We then examine the effect of regularization when applied to two key enablers of SAM: (a) the contextualized embedding layers, for their ability to capture well-scaled representations with respect to the values, and (b) the prescaled attention maps, for carrying value-independent global contextual information. We depict the perks of each distilling strategy on two image recognition benchmarks (CIFAR100 and ImageNet-32) – while (a) leads to a better overall accuracy, (b) helps enhance the rigidity by maintaining competitive performances. Furthermore, we identify the limitation imposed by the symmetric nature of regularization losses. To alleviate this, we propose an asymmetric variant and apply it to the pooled output distillation (POD) loss adapted for ViTs. Our experiments confirm that introducing asymmetry to POD boosts its plasticity while retaining stability across (a) and (b). Moreover, we acknowledge low forgetting measures for all the compared methods, indicating that ViTs might be naturally inclined continual learners. 1
Address
...
Corporate Author
...
Thesis
Bachelor's thesis
Master's thesis
Ph.D. thesis
Diploma thesis
Doctoral thesis
Habilitation thesis
Publisher
...
Place of Publication
...
Editor
...
Language
...
Summary Language
...
Original Title
...
Series Editor
...
Series Title
...
Abbreviated Series Title
...
Series Volume
...
Series Issue
...
Edition
...
ISSN
...
ISBN
...
Medium
...
Area
...
Expedition
...
Conference
...
Notes
...
Approved
yes
no
Location
Call Number
...
Serial
Marked
yes
no
Copy
true
fetch
ordered
false
Selected
yes
no
User Keys
...
User Notes
...
User File
...
User Groups
...
Cite Key
...
Related
...
File
URL
...
DOI
...
Online publication. Cite with this text:
...
Location Field:
don't touch
add
remove
my name & email address
Home
SQL Search
|
Library Search
|
Show Record
|
Extract Citations
Help