King's College London

Research portal

Stochastic filter groups for multi-task cnns: Learning specialist and generalist convolution kernels

Research output: Chapter in Book/Report/Conference proceedingConference paper

Felix Bragman, Ryutaro Tanno, Sebastien Ourselin, Daniel Alexander, Jorge Cardoso

Original languageEnglish
Title of host publicationProceedings - 2019 International Conference on Computer Vision, ICCV 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1385-1394
Number of pages10
ISBN (Electronic)9781728148038
DOIs
Publication statusPublished - Oct 2019
Event17th IEEE/CVF International Conference on Computer Vision, ICCV 2019 - Seoul, Korea, Republic of
Duration: 27 Oct 20192 Nov 2019

Publication series

NameProceedings of the IEEE International Conference on Computer Vision
Volume2019-October
ISSN (Print)1550-5499

Conference

Conference17th IEEE/CVF International Conference on Computer Vision, ICCV 2019
CountryKorea, Republic of
CitySeoul
Period27/10/20192/11/2019

King's Authors

Abstract

The performance of multi-task learning in Convolutional Neural Networks (CNNs) hinges on the design of feature sharing between tasks within the architecture. The number of possible sharing patterns are combinatorial in the depth of the network and the number of tasks, and thus hand-crafting an architecture, purely based on the human intuitions of task relationships can be time-consuming and suboptimal. In this paper, we present a probabilistic approach to learning task-specific and shared representations in CNNs for multi-task learning. Specifically, we propose 'stochastic filter groups' (SFG), a mechanism to assign convolution kernels in each layer to 'specialist' and 'generalist' groups, which are specific to and shared across different tasks, respectively. The SFG modules determine the connectivity between layers and the structures of task-specific and shared representations in the network. We employ variational inference to learn the posterior distribution over the possible grouping of kernels and network parameters. Experiments demonstrate the proposed method generalises across multiple tasks and shows improved performance over baseline methods.

View graph of relations

© 2018 King's College London | Strand | London WC2R 2LS | England | United Kingdom | Tel +44 (0)20 7836 5454