TY - GEN
T1 - Attention to Warp
T2 - 16th International Conference on Document Analysis and Recognition, ICDAR 2021
AU - Matsuo, Shinnosuke
AU - Wu, Xiaomeng
AU - Atarsaikhan, Gantugs
AU - Kimura, Akisato
AU - Kashino, Kunio
AU - Iwana, Brian Kenji
AU - Uchida, Seiichi
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - Deep time series metric learning is challenging due to the difficult trade-off between temporal invariance to nonlinear distortion and discriminative power in identifying non-matching sequences. This paper proposes a novel neural network-based approach for robust yet discriminative time series classification and verification. This approach adapts a parameterized attention model to time warping for greater and more adaptive temporal invariance. It is robust against not only local but also large global distortions, so that even matching pairs that do not satisfy the monotonicity, continuity, and boundary conditions can still be successfully identified. Learning of this model is further guided by dynamic time warping to impose temporal constraints for stabilized training and higher discriminative power. It can learn to augment the inter-class variation through warping, so that similar but different classes can be effectively distinguished. We experimentally demonstrate the superiority of the proposed approach over previous non-parametric and deep models by combining it with a deep online signature verification framework, after confirming its promising behavior in single-letter handwriting classification on the Unipen dataset.
AB - Deep time series metric learning is challenging due to the difficult trade-off between temporal invariance to nonlinear distortion and discriminative power in identifying non-matching sequences. This paper proposes a novel neural network-based approach for robust yet discriminative time series classification and verification. This approach adapts a parameterized attention model to time warping for greater and more adaptive temporal invariance. It is robust against not only local but also large global distortions, so that even matching pairs that do not satisfy the monotonicity, continuity, and boundary conditions can still be successfully identified. Learning of this model is further guided by dynamic time warping to impose temporal constraints for stabilized training and higher discriminative power. It can learn to augment the inter-class variation through warping, so that similar but different classes can be effectively distinguished. We experimentally demonstrate the superiority of the proposed approach over previous non-parametric and deep models by combining it with a deep online signature verification framework, after confirming its promising behavior in single-letter handwriting classification on the Unipen dataset.
UR - http://www.scopus.com/inward/record.url?scp=85115328215&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85115328215&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-86334-0_23
DO - 10.1007/978-3-030-86334-0_23
M3 - Conference contribution
AN - SCOPUS:85115328215
SN - 9783030863333
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 350
EP - 365
BT - Document Analysis and Recognition - ICDAR 2021 - 16th International Conference, Proceedings
A2 - Lladós, Josep
A2 - Lopresti, Daniel
A2 - Uchida, Seiichi
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 5 September 2021 through 10 September 2021
ER -