Change search
Link to record
Permanent link

Direct link
BETA
Li, Vladimir
Publications (2 of 2) Show all publications
Li, V. & Maki, A. (2018). Feature Contraction: New ConvNet Regularization in Image Classification. In: : . Paper presented at British Machine Vision Conference.
Open this publication in new window or tab >>Feature Contraction: New ConvNet Regularization in Image Classification
2018 (English)Conference paper, Oral presentation with published abstract (Refereed)
National Category
Computer Systems
Identifiers
urn:nbn:se:kth:diva-250575 (URN)
Conference
British Machine Vision Conference
Note

QC 20190624

Available from: 2019-04-30 Created: 2019-04-30 Last updated: 2019-06-24Bibliographically approved
Yang, Z., Li, V., Okada, R. & Maki, A. (2018). Target aware network adaptation for efficient representation learning. In: ECCV 2018: Computer Vision – ECCV 2018 Workshops: . Paper presented at 15th European Conference on Computer Vision, ECCV 2018; Munich; Germany; 8 September 2018 through 14 September 2018 (pp. 450-467). Munich: Springer, 11132
Open this publication in new window or tab >>Target aware network adaptation for efficient representation learning
2018 (English)In: ECCV 2018: Computer Vision – ECCV 2018 Workshops, Munich: Springer, 2018, Vol. 11132, p. 450-467Conference paper, Published paper (Refereed)
Abstract [en]

This paper presents an automatic network adaptation method that finds a ConvNet structure well-suited to a given target task, e.g. image classification, for efficiency as well as accuracy in transfer learning. We call the concept target-aware transfer learning. Given only small-scale labeled data, and starting from an ImageNet pre-trained network, we exploit a scheme of removing its potential redundancy for the target task through iterative operations of filter-wise pruning and network optimization. The basic motivation is that compact networks are on one hand more efficient and should also be more tolerant, being less complex, against the risk of overfitting which would hinder the generalization of learned representations in the context of transfer learning. Further, unlike existing methods involving network simplification, we also let the scheme identify redundant portions across the entire network, which automatically results in a network structure adapted to the task at hand. We achieve this with a few novel ideas: (i) cumulative sum of activation statistics for each layer, and (ii) a priority evaluation of pruning across multiple layers. Experimental results by the method on five datasets (Flower102, CUB200-2011, Dog120, MIT67, and Stanford40) show favorable accuracies over the related state-of-the-art techniques while enhancing the computational and storage efficiency of the transferred model.

Place, publisher, year, edition, pages
Munich: Springer, 2018
Series
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), ISSN 0302-9743 ; 11132
National Category
Computer Systems
Identifiers
urn:nbn:se:kth:diva-250561 (URN)10.1007/978-3-030-11018-5_38 (DOI)2-s2.0-85061697164 (Scopus ID)9783030110178 (ISBN)
Conference
15th European Conference on Computer Vision, ECCV 2018; Munich; Germany; 8 September 2018 through 14 September 2018
Note

QC 20190627

Available from: 2019-04-30 Created: 2019-04-30 Last updated: 2019-06-27Bibliographically approved
Organisations

Search in DiVA

Show all publications