Cross-view kernel transfer

Riikka Huusari; Cécile Capponi; Paul Villoutreix; Hachem Kadri

doi:10.1016/j.patcog.2022.108759

Cross-view kernel transfer

Riikka Huusari^*
, Cécile Capponi
, Paul Villoutreix
, Hachem Kadri

^*Corresponding author for this work

Research output: Contribution to journal › Article › Scientific › peer-review

62 Downloads (Pure)

Abstract

We consider the kernel completion problem with the presence of multiple views in the data. In this context the data samples can be fully missing in some views, creating missing columns and rows to the kernel matrices that are calculated individually for each view. We propose to solve the problem of completing the kernel matrices with Cross-View Kernel Transfer (CVKT) procedure, in which the features of the other views are transformed to represent the view under consideration. The transformations are learned with kernel alignment to the known part of the kernel matrix, allowing for finding generalizable structures in the kernel matrix under completion. Its missing values can then be predicted with the data available in other views. We illustrate the benefits of our approach with simulated data, multivariate digits dataset and multi-view dataset on gesture classification, as well as with real biological datasets from studies of pattern formation in early Drosophila melanogaster embryogenesis.

Original language	English
Article number	108759
Pages (from-to)	1-14
Number of pages	14
Journal	Pattern Recognition
Volume	129
DOIs	https://doi.org/10.1016/j.patcog.2022.108759
Publication status	Published - Sept 2022
MoE publication type	A1 Journal article-refereed

Funding

This work is mainly granted by the french national project ANR Lives ANR-15-CE23-0026, and by the Turing Center for Living Systems (CENTURI) for PV. For the most part work by RH has been done in Aix-Marseille University – the part in Aalto University has been funded by Academy of Finland grants 334790 (MAGITICS) and 310107 (MACOME).

Keywords

Cross-view transfer
Kernel completion
Kernel learning
Multi-view learning

Access to Document

10.1016/j.patcog.2022.108759Licence: CC BY

Cross-View kernel transferFinal published version, 3.5 MBLicence: CC BY

https://www.sciencedirect.com/science/article/pii/S0031320322002400

Cite this

@article{c460f9930a0b4a048c3d22683dda26cf,

title = "Cross-view kernel transfer",

abstract = "We consider the kernel completion problem with the presence of multiple views in the data. In this context the data samples can be fully missing in some views, creating missing columns and rows to the kernel matrices that are calculated individually for each view. We propose to solve the problem of completing the kernel matrices with Cross-View Kernel Transfer (CVKT) procedure, in which the features of the other views are transformed to represent the view under consideration. The transformations are learned with kernel alignment to the known part of the kernel matrix, allowing for finding generalizable structures in the kernel matrix under completion. Its missing values can then be predicted with the data available in other views. We illustrate the benefits of our approach with simulated data, multivariate digits dataset and multi-view dataset on gesture classification, as well as with real biological datasets from studies of pattern formation in early Drosophila melanogaster embryogenesis.",

keywords = "Cross-view transfer, Kernel completion, Kernel learning, Multi-view learning",

author = "Riikka Huusari and C{\'e}cile Capponi and Paul Villoutreix and Hachem Kadri",

note = "Funding Information: This work is mainly granted by the french national project ANR Lives ANR-15-CE23-0026, and by the Turing Center for Living Systems (CENTURI) for PV. For the most part work by RH has been done in Aix-Marseille University – the part in Aalto University has been funded by Academy of Finland grants 334790 (MAGITICS) and 310107 (MACOME). Publisher Copyright: {\textcopyright} 2022 The Author(s)",

year = "2022",

month = sep,

doi = "10.1016/j.patcog.2022.108759",

language = "English",

volume = "129",

pages = "1--14",

journal = "Pattern Recognition",

issn = "0031-3203",

publisher = "Elsevier",

}

TY - JOUR

T1 - Cross-view kernel transfer

AU - Huusari, Riikka

AU - Capponi, Cécile

AU - Villoutreix, Paul

AU - Kadri, Hachem

N1 - Funding Information: This work is mainly granted by the french national project ANR Lives ANR-15-CE23-0026, and by the Turing Center for Living Systems (CENTURI) for PV. For the most part work by RH has been done in Aix-Marseille University – the part in Aalto University has been funded by Academy of Finland grants 334790 (MAGITICS) and 310107 (MACOME). Publisher Copyright: © 2022 The Author(s)

PY - 2022/9

Y1 - 2022/9

N2 - We consider the kernel completion problem with the presence of multiple views in the data. In this context the data samples can be fully missing in some views, creating missing columns and rows to the kernel matrices that are calculated individually for each view. We propose to solve the problem of completing the kernel matrices with Cross-View Kernel Transfer (CVKT) procedure, in which the features of the other views are transformed to represent the view under consideration. The transformations are learned with kernel alignment to the known part of the kernel matrix, allowing for finding generalizable structures in the kernel matrix under completion. Its missing values can then be predicted with the data available in other views. We illustrate the benefits of our approach with simulated data, multivariate digits dataset and multi-view dataset on gesture classification, as well as with real biological datasets from studies of pattern formation in early Drosophila melanogaster embryogenesis.

AB - We consider the kernel completion problem with the presence of multiple views in the data. In this context the data samples can be fully missing in some views, creating missing columns and rows to the kernel matrices that are calculated individually for each view. We propose to solve the problem of completing the kernel matrices with Cross-View Kernel Transfer (CVKT) procedure, in which the features of the other views are transformed to represent the view under consideration. The transformations are learned with kernel alignment to the known part of the kernel matrix, allowing for finding generalizable structures in the kernel matrix under completion. Its missing values can then be predicted with the data available in other views. We illustrate the benefits of our approach with simulated data, multivariate digits dataset and multi-view dataset on gesture classification, as well as with real biological datasets from studies of pattern formation in early Drosophila melanogaster embryogenesis.

KW - Cross-view transfer

KW - Kernel completion

KW - Kernel learning

KW - Multi-view learning

UR - https://www.scopus.com/pages/publications/85129746995

U2 - 10.1016/j.patcog.2022.108759

DO - 10.1016/j.patcog.2022.108759

M3 - Article

AN - SCOPUS:85129746995

SN - 0031-3203

VL - 129

SP - 1

EP - 14

JO - Pattern Recognition

JF - Pattern Recognition

M1 - 108759

ER -

Cross-view kernel transfer

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

MAGITICS: Machine Learning for Antimicrobial Resistance

Machine Learning for Computational Metabolomics

Science-IT

Cite this

Cross-view kernel transfer

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Projects

MAGITICS: Machine Learning for Antimicrobial Resistance

Machine Learning for Computational Metabolomics

Equipment

Science-IT

Cite this