Deep Canonical Time Warping for simultaneous alignment and representation learning of sequences
- Submitting institution
-
Goldsmiths' College
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 2010
- Type
- D - Journal article
- DOI
-
10.1109/TPAMI.2017.2710047
- Title of journal
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Article number
- -
- First page
- 1128
- Volume
- 40
- Issue
- 5
- ISSN
- 0162-8828
- Open access status
- Compliant
- Month of publication
- May
- Year of publication
- 2017
- URL
-
http://research.gold.ac.uk/id/eprint/20538/
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
3
- Research group(s)
-
-
- Citation count
- 18
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- The authors propose learning hierarchical, shared representations via spatial and temporal non-linear transformations on high-dimensional data. By optimizing the proposed loss function on raw signals, it outperforms state-of-the-art methods in problems such as the alignment of (i) of acoustic and visual data, (ii) human actions, (iii) facial muscles (action units), and (iv) acoustic and articulatory recordings (where we achieve an order of magnitude better results against compared methods). This work has been presented at CVPR conference and as an invited talk at the Matrix and Tensor Factorization Methods in Computer Vision workshop at ICCV conference.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -