BIR: A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External Variables

Rebecca Marion; Adrien Bibal; Benoît Frénay

doi:10.1016/j.neucom.2018.11.093

BIR: A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External Variables

Rebecca Marion, Adrien Bibal, Benoît Frénay

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

37 Downloads (Pure)

Abstract

Interpreting nonlinear dimensionality reduction models using external features (or external variables) is crucial in many fields, such as psychology and ecology. Multidimensional scaling (MDS) is one of the most frequently used dimensionality reduction techniques in these fields. However, the rotation invariance of the MDS objective function may make interpretation of the resulting embedding difficult. This paper analyzes how the rotation of MDS embeddings affects sparse regression models used to interpret them and proposes a method, called the Best Interpretable Rotation (BIR) method, which selects the best MDS rotation for interpreting embeddings using external information.

Original language	English
Pages (from-to)	83-96
Number of pages	14
Journal	Neurocomputing
Volume	342
DOIs	https://doi.org/10.1016/j.neucom.2018.11.093
Publication status	Published - 4 Feb 2019

Keywords

Machine learning
Interpretability
Dimensionality reduction
Multidimensional scaling
Orthogonal transformation
Multi-view
Sparsity
Lasso regularization

Access to Document

10.1016/j.neucom.2018.11.093

2019-Marion_Bibal-BIR- A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External VariablesSubmitted manuscript, 2.49 MB

Cite this

@article{51f08cd4e241408c9340f527dd2779f1,

title = "BIR: A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External Variables",

abstract = "Interpreting nonlinear dimensionality reduction models using external features (or external variables) is crucial in many fields, such as psychology and ecology. Multidimensional scaling (MDS) is one of the most frequently used dimensionality reduction techniques in these fields. However, the rotation invariance of the MDS objective function may make interpretation of the resulting embedding difficult. This paper analyzes how the rotation of MDS embeddings affects sparse regression models used to interpret them and proposes a method, called the Best Interpretable Rotation (BIR) method, which selects the best MDS rotation for interpreting embeddings using external information.",

keywords = "Machine learning, Interpretability, Dimensionality reduction, Multidimensional scaling, Orthogonal transformation, Multi-view, Sparsity, Lasso regularization",

author = "Rebecca Marion and Adrien Bibal and Beno{\^i}t Fr{\'e}nay",

note = "Funding Information: The authors would like to thank Nathan Nguyen from the Universit{\'e} catholique de Louvain for having pointed to the need for this kind of method in psychology, as well as Prof. Bernadette Govaerts and Prof. Rainer von Sachs from the Universit{\'e} catholique de Louvain for their insights on the subject. The first author gratefully acknowledges financial support from the Belgian Fund for Scientific Research (F.R.S.-FNRS, FRIA grant). Funding Information: The authors would like to thank Nathan Nguyen from the Universit? catholique de Louvain for having pointed to the need for this kind of method in psychology, as well as Prof. Bernadette Govaerts and Prof. Rainer von Sachs from the Universit? catholique de Louvain for their insights on the subject. The first author gratefully acknowledges financial support from the Belgian Fund for Scientific Research (F.R.S.-FNRS, FRIA grant). Publisher Copyright: {\textcopyright} 2019 Elsevier B.V.",

year = "2019",

month = feb,

day = "4",

doi = "10.1016/j.neucom.2018.11.093",

language = "English",

volume = "342",

pages = "83--96",

journal = "Neurocomputing",

issn = "0925-2312",

publisher = "Elsevier",

}

TY - JOUR

T1 - BIR: A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External Variables

AU - Marion, Rebecca

AU - Bibal, Adrien

AU - Frénay, Benoît

N1 - Funding Information: The authors would like to thank Nathan Nguyen from the Université catholique de Louvain for having pointed to the need for this kind of method in psychology, as well as Prof. Bernadette Govaerts and Prof. Rainer von Sachs from the Université catholique de Louvain for their insights on the subject. The first author gratefully acknowledges financial support from the Belgian Fund for Scientific Research (F.R.S.-FNRS, FRIA grant). Funding Information: The authors would like to thank Nathan Nguyen from the Universit? catholique de Louvain for having pointed to the need for this kind of method in psychology, as well as Prof. Bernadette Govaerts and Prof. Rainer von Sachs from the Universit? catholique de Louvain for their insights on the subject. The first author gratefully acknowledges financial support from the Belgian Fund for Scientific Research (F.R.S.-FNRS, FRIA grant). Publisher Copyright: © 2019 Elsevier B.V.

PY - 2019/2/4

Y1 - 2019/2/4

N2 - Interpreting nonlinear dimensionality reduction models using external features (or external variables) is crucial in many fields, such as psychology and ecology. Multidimensional scaling (MDS) is one of the most frequently used dimensionality reduction techniques in these fields. However, the rotation invariance of the MDS objective function may make interpretation of the resulting embedding difficult. This paper analyzes how the rotation of MDS embeddings affects sparse regression models used to interpret them and proposes a method, called the Best Interpretable Rotation (BIR) method, which selects the best MDS rotation for interpreting embeddings using external information.

AB - Interpreting nonlinear dimensionality reduction models using external features (or external variables) is crucial in many fields, such as psychology and ecology. Multidimensional scaling (MDS) is one of the most frequently used dimensionality reduction techniques in these fields. However, the rotation invariance of the MDS objective function may make interpretation of the resulting embedding difficult. This paper analyzes how the rotation of MDS embeddings affects sparse regression models used to interpret them and proposes a method, called the Best Interpretable Rotation (BIR) method, which selects the best MDS rotation for interpreting embeddings using external information.

KW - Machine learning

KW - Interpretability

KW - Dimensionality reduction

KW - Multidimensional scaling

KW - Orthogonal transformation

KW - Multi-view

KW - Sparsity

KW - Lasso regularization

UR - https://www.sciencedirect.com/science/article/pii/S0925231219301481

UR - http://www.scopus.com/inward/record.url?scp=85061815947&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2018.11.093

DO - 10.1016/j.neucom.2018.11.093

M3 - Article

SN - 0925-2312

VL - 342

SP - 83

EP - 96

JO - Neurocomputing

JF - Neurocomputing

ER -

BIR: A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External Variables

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Interpretability and Explainability in Machine Learning and their Application to Nonlinear Dimensionality Reduction

Cite this

BIR: A Method for Selecting the Best Interpretable Multidimensional Scaling Rotation using External Variables

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Student theses

Interpretability and Explainability in Machine Learning and their Application to Nonlinear Dimensionality Reduction

Cite this