Latent Variable Analysis and Signal Separation

Name: Latent Variable Analysis and Signal Separation | 13th International Conference, LVA/ICA 2017, Grenoble, France, February 21-23, 2017, Proceedings
Brand: Springer
Price: 53.49 EUR
Availability: OnlineOnly

13th International Conference, LVA/ICA 2017, Grenoble, France, February 21-23, 2017, Proceedings

Petr Tichavský Massoud Babaie-Zadeh Olivier J.J. Michel Nadège Thirion-Moreau(Editor)

Springer (Publisher)

Published on 13. February 2017

XVI, 576 pages

E-Book

PDF with digital watermarking

System requirements

978-3-319-53547-0 (ISBN)

€53.49incl. 7% vat

System requirements

for PDF with digital watermarking

E-Book Single Licence

Available for download

Description

More details

Other editions

Content

Intro
Preface
Organization
Contents
Tensor Approaches
Higher-Order Block Term Decomposition for Spatially Folded fMRI Data
1 Introduction
1.1 Notation
2 Tensorial fMRI Analysis
2.1 Canonical Polyadic Decomposition (CPD)
2.2 Tensor Probabilistic Independent Component Analysis (TPICA)
3 Block Term Decomposition (BTD) for fMRI
3.1 Uniqueness
4 Simulation Results
4.1 Simulation of a Perception Study
4.2 Multi-slice Simulation
5 Conclusions
References
Modeling Parallel Wiener-Hammerstein Systems Using Tensor Decomposition of Volterra Kernels
1 Introduction
2 Volterra Kernels, Tensors and Tensor Decomposition
2.1 The Volterra Model for Nonlinear Systems
2.2 From Polynomials to Tensors
2.3 Canonical Polyadic Decomposition
3 Parallel Wiener-Hammerstein as Tensor Decomposition
3.1 Wiener-Hammerstein as Structured Tensor Decomposition
3.2 Parallel Wiener-Hammerstein Structure
3.3 Coupled Tensor and Matrix Decompositions
4 Numerical Results
5 Conclusions
References
Fast Nonnegative Matrix Factorization and Completion Using Nesterov Iterations
1 Introduction
2 Standard NeNMF
3 Extending NeNMF to Missing Entries
3.1 Weigthed Extension of NeNMF
3.2 EM Extension of NeNMF
4 Performance of the Weighted NeNMF Extensions
5 Conclusion
A Proof of Lemmas 1 and 2
References
Blind Source Separation of Single Channel Mixture Using Tensorization and Tensor Diagonalization
1 Introduction
2 General Framework for BSS of Single Channel Mixture
3 Tensorization of Sinusoid Signals
3.1 Two-Way and Three-Way Foldings
3.2 Toeplitzation
4 Simulations
5 Conclusions
A Appendix: Low-Rank Representation of the Sequence x(t) = tn
References
High-Resolution Subspace-Based Methods: Eigenvalue- or Eigenvector-Based Estimation?
1 Introduction
2 Multilevel Hankel Matrices and Their Subspaces
2.1 Definition and Factorization
2.2 Shift Properties of Subspaces
3 ESPRIT-Type Algorithms for MH Matrices
3.1 N-D ESPRIT Algorithm
3.2 IMDF Algorithm
3.3 IMDF Based on Least Squares (IMDF LS)
4 Perturbation Analysis
4.1 Basic Expressions
4.2 IMDF Perturbations
4.3 IMDF LS Perturbations
4.4 Computing the First-Order Perturbation and Its Moments
5 Simulations
6 Conclusions
References
From Source Positions to Room Properties: Learning Methods for Audio Scene Geometry Estimation
Speaker Tracking on Multiple-Manifolds with Distributed Microphones
1 Introduction
2 Problem Formulation
3 Multiple-Manifold Gaussian Process
4 Multiple-Manifold Speaker Tracking
5 Experimental Study
6 Conclusions
References
VAST: The Virtual Acoustic Space Traveler Dataset
1 Introduction
2 Dataset Design
2.1 General Principles
2.2 Room Simulation and Data Generation
2.3 Room Properties: Size and Surfaces
2.4 Reverberation Time
2.5 Source and Receiver Positions
2.6 Test Sets
3 Virtually Supervised Sound Source Localization
4 Conclusion
References
Sketching for Nearfield Acoustic Imaging of Heavy-Tailed Sources
1 Introduction
2 Mixture Model and -Stable Theory
2.1 Notation and Convolutive Model
2.2 Independent Isotropic -Stable Model for the Sources
2.3 The Levy Exponent and the Spatial Measure
3 Parameter Estimation
3.1 Sketching for the Levy Exponent
3.2 A Proposed NMF Algorithm to Determine
4 Evaluation
5 Conclusion
References
Acoustic DoA Estimation by One Unsophisticated Sensor
1 Introduction
2 Localization of Noise Sources
2.1 Geometrical Structure
2.2 Structure Quality
2.3 Conditions for Localization
3 Algorithms
3.1 Subspace Model
3.2 Dictionary Model
4 Numerical Results
4.1 White Sources
4.2 Speech Sources
5 Conclusion
References
Acoustic Source Localization by Combination of Supervised Direction-of-Arrival Estimation with Disjoint Component Analysis
1 Introduction
2 Methods
2.1 Probabilistic Source Localization
2.2 Disjoint Component Analysis
2.3 Decomposition of Source Probability Map
2.4 Multi-channel Signal Enhancement
3 Experiments and Results
4 Summary and Discussion
References
Tensors and Audio
An Initialization Method for Nonlinear Model Reduction Using the CP Decomposition
1 Introduction
2 Notations and Problem Statement
3 Finding an Appropriate Initialization
4 Simulations and Results
5 Case Study
6 Conclusion
References
Audio Zoom for Smartphones Based on Multiple Adaptive Beamformers
1 Introduction
2 Proposed Audio Zoom System
2.1 Target Sound Source Enhancement
2.2 Proposed Audio Zoom Effect Creation
3 Experiments
3.1 Experiment Setup
3.2 Result with Subjective Test
4 Conclusion
References
Complex Valued Robust Multidimensional SOBI
1 Introduction
2 Preliminaries
3 Algorithms
3.1 SOBI
3.2 Affine Equivariant SAM-SOBI
3.3 Multidimensional SOBI
3.4 Robust Multidimensional SOBI
4 Simulation Study
5 Conclusions
References
Ego Noise Reduction for Hose-Shaped Rescue Robot Combining Independent Low-Rank Matrix Analysis and Multichannel Noise Cancellation
1 Introduction
2 Hose-Shaped Rescue Robot and Ego Noise
2.1 Hose-Shaped Rescue Robot
2.2 Problem in Recording Speech
3 Overview of Independent Low-Rank Matrix Analysis
3.1 Formulation
3.2 Independent Low-Rank Matrix Analysis
4 Multichannel Noise Canceller
4.1 Conventional Method
4.2 Proposed Method
4.3 Flow of the Proposed Method
5 Experiment
5.1 Conditions
5.2 Results
6 Conclusion
References
Some Theory on Non-negative Tucker Decomposition
1 Tensor Decomposition Models
1.1 Tucker Decompositions
1.2 Canonical Polyadic Decomposition
2 Propagating Non-negativity and Non-negative Rank Through NTD
2.1 Elements of Cone Theory
2.2 Working Hypotheses
2.3 Propagating the Non-negative Rank to the Core
2.4 Propagating Non-negativity to the Core
3 Simulations
3.1 Some Algorithms for NTD and NMF
3.2 Some Tests on the Outputs of Algorithms
4 Conclusion
References
A New Algorithm for Multimodal Soft Coupling
1 Introduction
2 Soft Coupling for NMF
2.1 NMF Model
2.2 Coupled NMF
3 The Proposed Algorithm
3.1 Update Rule for Updating H1
3.2 Update Rule for Updating
4 Experimental Results
5 Conclusion
References
Adaptive Blind Separation of Instantaneous Linear Mixtures of Independent Sources
1 Introduction
2 Signal Model and Separation Performance Limits
3 Adaptive BSS Algorithms
3.1 Scaled Stochastic Natural Gradient Algorithm
3.2 Adaptive BGSEP
3.3 Adaptive BARBI
4 Experiments
5 Conclusion
References
Source Separation, Dereverberation and Noise Reduction Using LCMV Beamformer and Postfilter
1 Introduction
2 Problem Formulation
3 Optimal Multichannel Speaker Separation, Dereverberation and Noise Reduction
4 Estimation of the Late Reverberation PSD Matrix
4.1 Estimator Based on a Temporal Model
4.2 Estimator Based on a Spatial Model
5 Performance Evaluation
5.1 Setup
5.2 Results
6 Conclusions
References
Toward Rank Disaggregation: An Approach Based on Linear Programming and Latent Variable Analysis
1 Introduction
2 Rank Aggregation
2.1 Rank-Aggregation as an Optimization Problem
2.2 Criteria for Rank Aggregation
2.3 Numerical Example: CAC 40 Ranking of the Top 10 French Companies
3 Towards Rank Disaggregation
3.1 Rank Disaggregation via a Multivariate Decomposition Approach
3.2 Numerical Experiment
4 Conclusion
References
A Proximal Approach for Nonnegative Tensor Decomposition
1 Introduction
2 Canonical Polyadic Decomposition of N-th Order Tensors
2.1 Model
2.2 Objective
3 Optimization Problem and Proximal Algorithm
3.1 Criterion Formulation, Assumptions and Properties
3.2 Proposed Algorithm
3.3 Criterion Choice: Related Gradient and Proximity Operators
4 Numerical Simulations: Application to 4-th Order CPD
5 Conclusion
References
Psychophysical Evaluation of Audio Source Separation Methods
Abstract
1 Introduction
2 Method
3 Results: Analysis I - Psychophysical Correlation
4 Results: Analysis II - Comparison of Separation Methods
5 Conclusion and Discussion
Acknowledgment
References
Audio Signal Processing
On the Use of Latent Mixing Filters in Audio Source Separation
1 Introduction
2 Latent Mixing Filters and Estimation of Source Image
2.1 Principle
2.2 General Expression of the Source Image MMSE Estimator
2.3 The Gaussian Case
2.4 Inference of Source Image Using Metropolis Algorithm
3 Experiments
4 Conclusion
References
Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks
1 Introduction
2 Problem Formulation of Audio SCSS
3 DNNs for source separation and enhancement
3.1 Training DNN-A for Source Separation
3.2 Training DNN-B for Discriminative Enhancement
3.3 Testing DNN-A and DNN-B
4 Experiments and Discussion
5 Conclusion
References
Audiovisual Speech Separation Based on Independent Vector Analysis Using a Visual Voice Activity Detector
1 Introduction
2 Mathematical Preliminaries
2.1 Notations
2.2 Mixing and Separation Models
3 Method
3.1 Cost Function and Learning Algorithm
3.2 Choice of the Source Prior
3.3 Integration of Activity Information
4 Experimental Results
4.1 Performance Measure
4.2 Experiments
4.3 Discussion
5 Conclusion
References
Monoaural Audio Source Separation Using Deep Convolutional Neural Networks
1 Introduction
2 Proposed Framework
2.1 Model Architecture
2.2 Time-Frequency Masking
2.3 Parameter Learning
3 Evaluation
3.1 Dataset
3.2 Adjustments to Learning Objective
3.3 Evaluation Setup
3.4 Experiments
4 Conclusions and Future Work
References
Theoretical Developments
On the Behaviour of the Estimated Fourth-Order Cumulants Matrix of a High-Dimensional Gaussian White Noise
1 Introduction
2 Presentation of the Problem
3 Conditions Under Which "026B30D N "026B30D 0
4 Study of the Case Where N and M2 Are of the Same Order of Magnitude
5 Concluding Remarks
References
Caveats with Stochastic Gradient and Maximum Likelihood Based ICA for EEG
1 Introduction
2 Infomax: Description of the Optimization Algorithm
3 Assessing the Performance of ICA Using Dipolarity
4 Numerical Experiments
5 Conclusion
References
Approximate Joint Diagonalization According to the Natural Riemannian Distance
1 Introduction
2 Method: Riemannian Distance and Optimization
2.1 Cone of Hermitian Positive Definite Matrices
2.2 Riemannian Diagonality Measure
2.3 AJD Based on the Riemannian Distance of H++n
2.4 Riemannian Optimization for the Subproblem
3 Numerical Experiments
3.1 Simulated Data
3.2 Electroencephalographic (EEG) Data
4 Conclusions
References
Gaussian Processes for Source Separation in Overdetermined Bilinear Mixtures
1 Introduction
2 Problem Statement
2.1 Bilinear Mixtures - The Overdetermined Case
3 Gaussian Processes in the Bilinear Mixtures
3.1 The Predictive Distribution
3.2 Maximization of the Marginal Likelihood
4 Simulation Results
5 Conclusions
References
Model-Independent Method of Nonlinear Blind Source Separation
1 Introduction
2 Method
3 Experiments
4 Conclusion
References
Physics and Bio Signal Processing
The 2016 Signal Separation Evaluation Campaign
1 Introduction
2 UND: Underdetermined-Speech and Music Mixtures
3 BGN: Two-Channel Mixtures of Speech and Real-World Background Noise
4 BIO: Separation of Biomedical Signals
5 MUS: Professionally-Produced Music Recordings
6 Conclusion
References
Multimodality for Rainfall Measurement
Abstract
1 Introduction
2 How Multimodal Measurements Are Currently Used: Calibration and Data Assimilation
3 Parametric and Non-parametric Data Fusion Information
4 Discussion and Conclusions
Acknowledgments
References
Particle Flow SMC-PHD Filter for Audio-Visual Multi-speaker Tracking
1 Introduction
2 AV-SMC-PHD Filter and Particle Flow
2.1 AV-SMC-PHD Filter
2.2 Particle Flow
3 Proposed AV-PF-SMC-PHD Filter
4 Experimental Results
5 Conclusion
References
Latent Variable Analysis in Observation Sciences
Estimation of the Intrinsic Dimensionality in Hyperspectral Imagery via the Hubness Phenomenon
1 Introduction
2 The Hubness Phenomenon
3 Properties of Hubness in Hyperspectral Data Sets
4 The Algorithm
5 Results
5.1 Artificial Data Sets
5.2 Real Data Sets
6 Conclusions
References
A Blind Identification and Source Separation Method Based on Subspace Intersections for Hyperspectral Astrophysical Data
1 Introduction
2 Problem Statement
2.1 Data Model
2.2 Geometric Framework
3 BSS Method Based on Subspace Intersection
3.1 Source Number Estimation
3.2 Pairs of Zones Identification
3.3 Mixing Matrix Estimation
3.4 Source Matrix Reconstruction
4 Experimental Results
4.1 Synthetic Data
4.2 Real Data
5 Conclusion and Future Work
References
Estimating the Number of Endmembers to Use in Spectral Unmixing of Hyperspectral Data with Collaborative Sparsity
1 Introduction
2 Proposed Approach
2.1 Extended Linear Mixing Model
2.2 Collaborative Sparsity for Hyperspectral Unmixing
2.3 Computing a Regularization Path
2.4 Selecting the Best Model
3 Results
3.1 Results on Simulated Data
3.2 Results on Real Data
4 Conclusion
References
Sharpening Hyperspectral Images Using Plug-and-Play Priors
1 Introduction
2 Problem Formulation
3 Optimization
4 Plugging a Gaussian Mixture Model Denoiser
5 Experimental Results
5.1 Denoising
5.2 Fusion
6 Conclusions and Future Work
References
On Extracting the Cosmic Microwave Background from Multi-channel Measurements
1 Introduction: Planck Data and the CMB
2 Cleaning Coherent Contamination, ICA Style
3 Maximum Likelihood Solutions
4 Discussion
5 Conclusion: What's the Point?
References
ICA Theory and Applications
Kernel-Based NPLS for Continuous Trajectory Decoding from ECoG Data for BCI Applications
Abstract
1 Introduction
2 Methods
2.1 Generic N-way PLS
2.2 NPLS with Kernel-Based Nonlinear Mapping
2.3 Criteria
3 Data Description
4 Results
5 Discussion and Conclusion
Acknowledgments
References
On the Optimal Non-linearities for Gaussian Mixtures in FastICA
1 Introduction
2 Optimal Non-linearities for Gaussian Mixtures
2.1 Gaussian Location Mixtures
2.2 Gaussian Scale Mixtures
3 Simulations
3.1 The Choice of the Tuning Parameter in tanh(ax)
3.2 Estimating Scale Mixtures and Heavy-Tailed Components
4 Discussion
References
Fast Disentanglement-Based Blind Quantum Source Separation and Process Tomography: A Closed-Form Solution Using a Feedback Classical Adapting Structure
1 Prior Work and Problem Statement
2 Mixing Model
3 Separating System, Separation Principle and Criterion
3.1 Inverting Block of Separating System
3.2 Adapting Block, Separation Principle and Criterion
4 Separation Algorithms
5 Blind Quantum Process Tomography
6 Conclusion
References
Blind Separation of Cyclostationary Sources with Common Cyclic Frequencies
1 Introduction
2 Problem Formulation
3 Proposed Method
3.1 Preliminaries
3.2 Construction of the Matrices Set
3.3 Non-unitary Joint Diagonalization Algorithm
3.4 Summary of the Proposed Method
4 Numerical Simulations
5 Conclusion
References
Adaptation of a Gaussian Mixture Regressor to a New Input Distribution: Extending the C-GMR Framework
1 Introduction
2 Cascaded GMR
2.1 Definitions, Notations and Working Hypothesis
2.2 D-GMR, SC-GMR and IC-GMR
3 Joint GMR
4 EM Algorithm for J-GMR
4.1 E-step
4.2 M-step
4.3 EM Initialization
4.4 A Remark on the Link Between the J-GMR EM and the IC-GMR EM
5 Experiments
6 Conclusion
References
Efficient Optimization of the Adaptive ICA Function with Estimating the Number of Non-Gaussian Sources
1 Introduction
2 Objective Function
3 Method
3.1 Optimization for Each Component
3.2 Derivation of a Threshold by the Fisher Information
3.3 Complete Algorithm
4 Results
5 Conclusion
References
Feasibility of WiFi Site-Surveying Using Crowdsourced Data
1 Introduction
2 Problem Statement
3 Modeling
3.1 Pedestrian Modeling
3.2 RSSI Map
4 Surveying Algorithm
4.1 Admissible Trajectories
4.2 Admissible Measurements
4.3 RSSI Map Update
4.4 PDR Refinement
5 Simulation Results
5.1 Case Study
5.2 Surveying Feasibility
6 Conclusions
References
On Minimum Entropy Deconvolution of Bi-level Images
1 Introduction
2 Image Deconvolution
3 Minimum Entropy Deconvolution
3.1 Gradient Based Algorithm
4 Blind Deconvolution of QR Codes
5 Simulation Results
5.1 Synthetic Scenario
5.2 Real Scenario
6 Conclusions
References
A Joint Second-Order Statistics and Density Matching-Based Approach for Separation of Post-Nonlinear Mixtures
1 Introduction
2 The Post-Nonlinear Mixtures
2.1 Separation Techniques for PNL Mixtures
3 Proposed Separation Method
3.1 Second-Order Statistics for Blind Separation
3.2 Matching of Gaussian Distributions
3.3 The Combined Approach
4 Simulation Results
5 Conclusions
References
Optimal Measurement Times for Observing a Brownian Motion over a Finite Period Using a Kalman Filter
1 Introduction
2 Model Description
3 Controlling the Maximal Variance
4 Controlling the Mean Variance
4.1 The Optimal Instant in Case of One Measure: Qualitative Results
4.2 Derivation and Properties of (16)
5 Conclusion and Perspectives
References
On Disjoint Component Analysis
1 Introduction
2 Disjoint Component Analysis
3 Analysis of the Minima of the Criterion
4 An Algorithm Based on Givens Rotations
5 Simulation Results
6 Conclusions
References
Sparsity-Aware Signal Processing
Accelerated Dictionary Learning for Sparse Signal Representation
1 Introduction
2 Proposed Method
2.1 Main Problem
2.2 Sparse Representation
2.3 Dictionary Update
2.4 Non-zero Coefficients Update
3 Simulations
4 Conclusion
References
BSS with Corrupted Data in Transformed Domains
1 Introduction
2 Sparsity and Morphological Diversity
3 The Algorithm rGMCA
3.1 Estimation of A and S
3.2 Estimation of S and O
3.3 Choice of the Parameters
4 Numerical Experiments
4.1 Monte Carlo Simulations
4.2 2D Simulations
5 Conclusion
References
Singing Voice Separation Using RPCA with Weighted l1-norm
1 Introduction
2 Algorithm
2.1 Robust Principal Component Analysis
2.2 RPCA with Weighted l1-norm
3 Singing Voice Separation
3.1 SVS Using RPCA
3.2 Proposed Method: SVS Using wRPCA
4 Experimental Results
5 Discussion
6 Conclusion
References
Multimodal Approach to Remove Ocular Artifacts from EEG Signals Using Multiple Measurement Vectors
1 Introduction
2 Proposed Method
2.1 Multiple Measurement Vectors
2.2 Data Matrix Building
2.3 Dictionary Selection
3 Experiments
3.1 MMV-G&E Parameters and Qualitative Results
3.2 Comparisons and Validation
4 Conclusions and Perspectives
References
Author Index

System requirements

Save as PDF Copy link into clipboard

Schweitzer Fachinformationen

Latent Variable Analysis and Signal Separation

Description

More details

Other editions

Additional editions

Content

System requirements