TY - CHAP
T1 - Stochastic filter groups for multi-task cnns
T2 - 17th IEEE/CVF International Conference on Computer Vision, ICCV 2019
AU - Bragman, Felix
AU - Tanno, Ryutaro
AU - Ourselin, Sebastien
AU - Alexander, Daniel
AU - Cardoso, Jorge
PY - 2019/10
Y1 - 2019/10
N2 - 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.
AB - 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.
UR - http://www.scopus.com/inward/record.url?scp=85081913659&partnerID=8YFLogxK
U2 - 10.1109/ICCV.2019.00147
DO - 10.1109/ICCV.2019.00147
M3 - Conference paper
AN - SCOPUS:85081913659
T3 - Proceedings of the IEEE International Conference on Computer Vision
SP - 1385
EP - 1394
BT - Proceedings - 2019 International Conference on Computer Vision, ICCV 2019
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 27 October 2019 through 2 November 2019
ER -