Consistent Spectral Methods for Dimensionality Reduction

Abstract : This paper addresses the problem of dimension reduction of noisy data, more precisely the challenge to determine the dimension of the subspace where the observed signal lives in. Based on results from random matrix theory, two novel estimators of the signal dimension are proposed in this paper. Consistency of the estimators is proved in the modern asymptotic regime, where the number of parameters grows proportionally with the sample size. Experimental results show that the novel estimators are robust to noise and, moreover, they give highly accurate results in settings where standard methods fail. We apply the novel dimension estimators to several life sciences benchmarks in the context of classification, and illustrate the improvements achieved by the new methods compared to the state-of-the-art approaches.
Document type :
Conference papers
Complete list of metadatas

https://hal-utt.archives-ouvertes.fr/hal-02307357
Contributor : Jean-Baptiste Vu Van <>
Submitted on : Monday, October 7, 2019 - 3:10:42 PM
Last modification on : Thursday, October 10, 2019 - 1:31:47 AM

Identifiers

Citation

Malika Kharouf, Tabea Rebafka, Nataliya Sokolovska. Consistent Spectral Methods for Dimensionality Reduction. 2018 26th European Signal Processing Conference (EUSIPCO), Sep 2018, Rome, Italy. pp.286-290, ⟨10.23919/EUSIPCO.2018.8553295⟩. ⟨hal-02307357⟩

Share

Metrics

Record views

9