Deep networks trained on large-scale data can learn transferable features to
promote learning multiple tasks. Since deep features eventually transition from
general to specific along deep networks, a fundamental problem of multi-task
learning is how to exploit the task relatedness underlying parameter tensors
and improve feature transferability in the multiple task-specific layers. This
paper presents Multilinear Relationship Networks (MRN) that discover the task
relationships based on novel tensor normal priors over parameter tensors of
multiple task-specific layers in deep convolutional networks. By jointly
learning transferable features and multilinear relationships of tasks and
features, MRN is able to alleviate the dilemma of negative-transfer in the
feature layers and under-transfer in the classifier layer. Experiments show
that MRN yields state-of-the-art results on three multi-task learning datasets.
Beschreibung
Learning Multiple Tasks with Multilinear Relationship Networks
%0 Generic
%1 long2015learning
%A Long, Mingsheng
%A Cao, Zhangjie
%A Wang, Jianmin
%A Yu, Philip S.
%D 2015
%K deep_learning multitask
%T Learning Multiple Tasks with Multilinear Relationship Networks
%U http://arxiv.org/abs/1506.02117
%X Deep networks trained on large-scale data can learn transferable features to
promote learning multiple tasks. Since deep features eventually transition from
general to specific along deep networks, a fundamental problem of multi-task
learning is how to exploit the task relatedness underlying parameter tensors
and improve feature transferability in the multiple task-specific layers. This
paper presents Multilinear Relationship Networks (MRN) that discover the task
relationships based on novel tensor normal priors over parameter tensors of
multiple task-specific layers in deep convolutional networks. By jointly
learning transferable features and multilinear relationships of tasks and
features, MRN is able to alleviate the dilemma of negative-transfer in the
feature layers and under-transfer in the classifier layer. Experiments show
that MRN yields state-of-the-art results on three multi-task learning datasets.
@misc{long2015learning,
abstract = {Deep networks trained on large-scale data can learn transferable features to
promote learning multiple tasks. Since deep features eventually transition from
general to specific along deep networks, a fundamental problem of multi-task
learning is how to exploit the task relatedness underlying parameter tensors
and improve feature transferability in the multiple task-specific layers. This
paper presents Multilinear Relationship Networks (MRN) that discover the task
relationships based on novel tensor normal priors over parameter tensors of
multiple task-specific layers in deep convolutional networks. By jointly
learning transferable features and multilinear relationships of tasks and
features, MRN is able to alleviate the dilemma of negative-transfer in the
feature layers and under-transfer in the classifier layer. Experiments show
that MRN yields state-of-the-art results on three multi-task learning datasets.},
added-at = {2018-04-27T11:39:46.000+0200},
author = {Long, Mingsheng and Cao, Zhangjie and Wang, Jianmin and Yu, Philip S.},
biburl = {https://www.bibsonomy.org/bibtex/2a956a6866e705126a2b6e8db1d45f314/dallmann},
description = {Learning Multiple Tasks with Multilinear Relationship Networks},
interhash = {2008d095f2c996b504395a12a2b74d71},
intrahash = {a956a6866e705126a2b6e8db1d45f314},
keywords = {deep_learning multitask},
note = {cite arxiv:1506.02117Comment: NIPS 2017},
timestamp = {2018-04-27T11:39:46.000+0200},
title = {Learning Multiple Tasks with Multilinear Relationship Networks},
url = {http://arxiv.org/abs/1506.02117},
year = 2015
}