A Comparative Study of Linear and Nonlinear Dimensionality Reduction for Speaker Identification

Errity, Andrew; McKenna, John

doi:10.1109/icdsp.2007.4288650

Cited by 6 publications

(3 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An affinity matrix was constructed with a normalized angle kernel, and eigenvectors were estimated via diffusion map embedding ( Figure 1A ), a nonlinear dimensionality reduction technique ( Coifman and Lafon, 2006 ) that projects connectome features into low-dimensional manifolds ( Margulies et al, 2016 ). This technique is only controlled by a few parameters, computationally efficient, and relatively robust to noise compared to other nonlinear techniques ( Errity and McKenna, 2007 ; Gallos et al, 2020 ; Hong et al, 2020 ; Tenenbaum et al, 2000 ), and has been extensively used in the previous gradient mapping literature ( Hong et al, 2019 ; Hong et al, 2020 ; Huntenburg et al, 2017 ; Larivière et al, 2020a ; Margulies et al, 2016 ; Müller et al, 2020 ; Paquola et al, 2019a ; Park et al, 2021b ; Valk et al, 2020 ; Vos de Wael et al, 2020a ). It is controlled by two parameters α and t , where α controls the influence of the density of sampling points on the manifold (α = 0, maximal influence; α = 1, no influence) and t controls the scale of eigenvalues of the diffusion operator.…”

Section: Methodsmentioning

confidence: 99%

An expanding manifold in transmodal regions characterizes adolescent reconfiguration of structural connectome organization

Park¹,

Bethlehem²,

Paquola³

et al. 2021

eLife

View full text Add to dashboard Cite

Adolescence is a critical time for the continued maturation of brain networks. Here, we assessed structural connectome development in a large longitudinal sample ranging from childhood to young adulthood. By projecting high-dimensional connectomes into compact manifold spaces, we identified a marked expansion of structural connectomes with the strongest effects in transmodal regions during adolescence. Findings reflected increased within-module connectivity together with increased segregation, indicating increasing differentiation of higher-order association networks from the rest of the brain. Projection of subcortico-cortical connectivity patterns into these manifolds showed parallel alterations in pathways centered on the caudate and thalamus. Connectome findings were contextualized via spatial transcriptome association analysis, highlighting genes enriched in cortex, thalamus, and striatum. Statistical learning of cortical and subcortical manifold features at baseline and their maturational change predicted measures of intelligence at follow-up. Our findings demonstrate that connectome manifold learning can bridge the conceptual and empirical gaps between macroscale network reconfigurations, microscale processes, and cognitive outcomes in adolescent development.

show abstract

Section: Methodsmentioning

confidence: 99%

An expanding manifold in transmodal regions characterizes adolescent reconfiguration of structural connectome organization

Park¹,

Bethlehem²,

Paquola³

et al. 2021

eLife

View full text Add to dashboard Cite

show abstract

“…1A ), a non-linear dimensionality reduction technique (Coifman and Lafon, 2006) that projects connectome features into low dimensional manifolds (Margulies et al, 2016). This technique is only controlled by few parameters, computationally efficient, and relatively robust to noise compared to other non-linear techniques (Errity and McKenna, 2007; Gallos et al, 2020; Hong et al, 2020; Tenenbaum et al, 2000), and has been extensively used in the previous gradient mapping literature (Hong et al, 2019, 2020; Huntenburg et al, 2017; Larivière et al, 2020a; Margulies et al, 2016; Müller et al, 2020; Paquola et al, 2019a; Park et al, 2021; Sofie L. Valk et al, 2020; Vos de Wael et al, 2020). It is controlled by two parameters α and t , where α controls the influence of the density of sampling points on the manifold (α = 0, maximal influence; α = 1, no influence) and t controls the scale of eigenvalues of the diffusion operator.…”

Section: Methodsmentioning

confidence: 99%

An expanding manifold in transmodal regions characterizes adolescent reconfiguration of structural connectome organization

Park

Bethlehem

Paquola

et al. 2020

Preprint

View full text Add to dashboard Cite

Adolescence is a critical time for the continued maturation of brain networks, and the current work assessed longitudinal reconfigurations of diffusion MRI derived connectomes in a large sample (n = 208). We identified an expansion of structural network representations in lower dimensional manifold spaces, with strongest effects in transmodal cortices and indicative of an increasing differentiation of these networks from the rest of the brain. Findings were shown to relate to mainly an increase in within-module connectivity and increased segregation during adolescence. Connectome findings were consistent when additionally controlling for regional changes in MRI measures of cortical morphology and myelin, despite reductions in effect sizes. On the other hand, cortical areas showing manifold expansion were found to become more closely embedded with subcortical targets, notably the striatum. Identified networks were also found to harbor genes significantly expressed in both cortical and subcortical regions. Using supervised machine learning with cross-validation, we demonstrated that cortico-subcortical manifold features at baseline and their maturational change predicted measures of intelligence at follow-up. Our findings chart adolescent connectome development and suggest that brain-wide network reconfigurations offer intermediary phenotypes between genetic-microstructural processes and cognitive-behavioral outcomes.We explored how whole-brain structural networks reconfigure during adolescence, using a novel analysis that simultaneously interrogates network and regional features. Our approach revealed an increasing differentiation of the connectome during this critical period, particularly in higher-order prefrontal and parietal regions. We could show that while this effect was only partially related to developmental trajectories of intracortical microstructure and morphology, but rather to changes in connectivity to subcortical areas, particularly the striatum. Cortico-subcortical network effects were also supported by transcriptomics and phenotype prediction, which suggested that cognitive function in adulthood was most effectively predicted when combining cortical and subcortical features. Our findings shed new light on adolescence and support a key role of cortico-subcortical integration in shaping macroscale structural networks. Park et al. | Structural connectome maturation during adolescence 3 3

show abstract

“…These manifold learning algorithms have been successfully applied to a number of speech processing applications including low dimensional visualization of speech, analysis of speech [10] and speaker recognition [11]. However, there is very little research on these manifold learning algorithms applied for phoneme recognition tasks.…”

Section: Introductionmentioning

confidence: 99%

Dimensionality reduction-based phoneme recognition

Zhang

Zhao

2008

2008 9th International Conference on Signal Processing

View full text Add to dashboard Cite

In this paper, linear and nonlinear dimensionality reduction algorithms are proposed to speech phoneme data from TIMIT corpus in an effort to perform dimensionality reduction for yielding low dimensional features capable of discriminating between phonemes. The linear dimensionality reduction method, including principal component analysis (PCA) and linear discriminant analysis (LDA), and the nonlinear dimensionality reduction method, including locally linear embedding (LLE) and isometric feature mapping (Isomap), are investigated. The resulting features by dimensionality reduction are evaluated in support vector machines (SVM)-based phoneme recognition experiments. Experiment results indicate that traditional linear LDA and PCA techniques for dimensionality reduction are capable of outperforming nonlinear LLE and Isomap techniques for phoneme recognition.

show abstract

A Comparative Study of Linear and Nonlinear Dimensionality Reduction for Speaker Identification

Cited by 6 publications

References 13 publications

An expanding manifold in transmodal regions characterizes adolescent reconfiguration of structural connectome organization

An expanding manifold in transmodal regions characterizes adolescent reconfiguration of structural connectome organization

An expanding manifold in transmodal regions characterizes adolescent reconfiguration of structural connectome organization

Dimensionality reduction-based phoneme recognition

Contact Info

Product

Resources

About