Visit ebookmass.

com to download the full version and

explore more ebook or textbook

Tensors for Data Processing. Theory, Methods, and

Applications Yipeng Liu

_____ Click the link below to download _____

https://ebookmass.com/product/tensors-for-data-processing-
theory-methods-and-applications-yipeng-liu/

Explore and download more ebook or textbook at ebookmass.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Tensors for Data Processing 1st Edition Yipeng Liu

https://ebookmass.com/product/tensors-for-data-processing-1st-edition-
yipeng-liu/

Spatial analysis using big data: methods and urban

applications Yamagata

https://ebookmass.com/product/spatial-analysis-using-big-data-methods-
and-urban-applications-yamagata/

Magnetic Communications: Theory and Techniques Liu

https://ebookmass.com/product/magnetic-communications-theory-and-
techniques-liu/

Computational Methods for Nonlinear Dynamical Systems:

Theory and Applications in Aerospace Engineering Xuechuan
Wang
https://ebookmass.com/product/computational-methods-for-nonlinear-
dynamical-systems-theory-and-applications-in-aerospace-engineering-
xuechuan-wang/
Processing for Android: Create Mobile, Sensor-aware, and
XR Applications Using Processing 2nd Edition Colubri

https://ebookmass.com/product/processing-for-android-create-mobile-
sensor-aware-and-xr-applications-using-processing-2nd-edition-colubri/

Big Data in Astronomy: Scientific Data Processing for

Advanced Radio Telescopes 1st Edition Linghe Kong (Editor)

https://ebookmass.com/product/big-data-in-astronomy-scientific-data-
processing-for-advanced-radio-telescopes-1st-edition-linghe-kong-
editor/

Oil Palm Biomass for Composite Panels: Fundamentals,

Processing, and Applications S.M. Sapuan

https://ebookmass.com/product/oil-palm-biomass-for-composite-panels-
fundamentals-processing-and-applications-s-m-sapuan/

Computational and Data-Driven Chemistry Using Artificial

Intelligence: Fundamentals, Methods and Applications
Takashiro Akitsu
https://ebookmass.com/product/computational-and-data-driven-chemistry-
using-artificial-intelligence-fundamentals-methods-and-applications-
takashiro-akitsu/

Processing for Android: Create Mobile, Sensor-aware, and

XR Applications Using Processing, 2nd Edition Andrés
Colubri
https://ebookmass.com/product/processing-for-android-create-mobile-
sensor-aware-and-xr-applications-using-processing-2nd-edition-andres-
colubri/
 Tensors for
Data Processing
This page intentionally left blank
 Tensors for
Data Processing
Theory, Methods, and Applications

Edited by
Yipeng Liu
School of Information and Communication Engineering
University of Electronic Science and Technology
of China (UESTC)
Chengdu, China
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
525 B Street, Suite 1650, San Diego, CA 92101, United States
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
Copyright © 2022 Elsevier Inc. All rights reserved.

MATLAB® is a trademark of The MathWorks, Inc. and is used with permission.

The MathWorks does not warrant the accuracy of the text or exercises in this book.
This book’s use or discussion of MATLAB® software or related products does not constitute
endorsement or sponsorship by The MathWorks of a particular pedagogical approach or particular use
of the MATLAB® software.
No part of this publication may be reproduced or transmitted in any form or by any means, electronic
or mechanical, including photocopying, recording, or any information storage and retrieval system,
without permission in writing from the publisher. Details on how to seek permission, further
information about the Publisher’s permissions policies and our arrangements with organizations such
as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website:
www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the
Publisher (other than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience
broaden our understanding, changes in research methods, professional practices, or medical treatment
may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating
and using any information, methods, compounds, or experiments described herein. In using such
information or methods they should be mindful of their own safety and the safety of others, including
parties for whom they have a professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume
any liability for any injury and/or damage to persons or property as a matter of products liability,
negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas
contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-824447-0

For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Mara Conner

Acquisitions Editor: Tim Pitts
Editorial Project Manager: Charlotte Rowley
Production Project Manager: Prem Kumar Kaliamoorthi
Designer: Miles Hitchen
Typeset by VTeX
Contents

List of contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
CHAPTER 1 Tensor decompositions: computations,
applications, and challenges . . . . . . . . . . . . . . . . . . . . 1
Yingyue Bi, Yingcong Lu, Zhen Long, Ce Zhu, and
Yipeng Liu
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 What is a tensor? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Why do we need tensors? . . . . . . . . . . . . . . . . . . . . . 2
1.2 Tensor operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Tensor notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Matrix operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Tensor transformations . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.4 Tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.5 Structural tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3 Tensor decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.1 Tucker decomposition . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.2 Canonical polyadic decomposition . . . . . . . . . . . . . . . 14
1.3.3 Block term decomposition . . . . . . . . . . . . . . . . . . . . . 16
1.3.4 Tensor singular value decomposition . . . . . . . . . . . . . 18
1.3.5 Tensor network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4 Tensor processing techniques . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.5 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
CHAPTER 2 Transform-based tensor singular value
decomposition in multidimensional image recovery 31
Tai-Xiang Jiang, Michael K. Ng, and Xi-Le Zhao
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2 Recent advances of the tensor singular value decomposition . . 34
2.2.1 Preliminaries and basic tensor notations . . . . . . . . . . . 34
2.2.2 The t-SVD framework . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2.3 Tensor nuclear norm and tensor recovery . . . . . . . . . . 38
2.2.4 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3 Transform-based t-SVD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.1 Linear invertible transform-based t-SVD . . . . . . . . . . 45

v
vi Contents

2.3.2 Beyond invertibility and data adaptivity . . . . . . . . . . . 47

2.4 Numerical experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.4.1 Examples within the t-SVD framework . . . . . . . . . . . 49
2.4.2 Examples of the transform-based t-SVD . . . . . . . . . . 51
2.5 Conclusions and new guidelines . . . . . . . . . . . . . . . . . . . . . . . 53
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
CHAPTER 3 Partensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Paris A. Karakasis, Christos Kolomvakis, George Lourakis,
George Lykoudis, Ioannis Marios Papagiannakos,
Ioanna Siaminou, Christos Tsalidis, and
Athanasios P. Liavas
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.1.1 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.1.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2 Tensor decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2.1 Matrix least-squares problems . . . . . . . . . . . . . . . . . . 65
3.2.2 Alternating optimization for tensor decomposition . . . 69
3.3 Tensor decomposition with missing elements . . . . . . . . . . . . . 70
3.3.1 Matrix least-squares with missing elements . . . . . . . . 71
3.3.2 Tensor decomposition with missing elements: the
unconstrained case . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.3 Tensor decomposition with missing elements: the
nonnegative case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.4 Alternating optimization for tensor decomposition
with missing elements . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4 Distributed memory implementations . . . . . . . . . . . . . . . . . . . 75
3.4.1 Some MPI preliminaries . . . . . . . . . . . . . . . . . . . . . . 75
3.4.2 Variable partitioning and data allocation . . . . . . . . . . . 77
3.4.3 Tensor decomposition . . . . . . . . . . . . . . . . . . . . . . . . 79
3.4.4 Tensor decomposition with missing elements . . . . . . . 81
3.4.5 Some implementation details . . . . . . . . . . . . . . . . . . . 82
3.5 Numerical experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5.1 Tensor decomposition . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5.2 Tensor decomposition with missing elements . . . . . . . 84
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
CHAPTER 4 A Riemannian approach to low-rank tensor
learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Hiroyuki Kasai, Pratik Jawanpuria, and Bamdev Mishra
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 A brief introduction to Riemannian optimization . . . . . . . . . . 93
 Contents vii

4.2.1 Riemannian manifolds . . . . . . . . . . . . . . . . . . . . . . . . 94

4.2.2 Riemannian quotient manifolds . . . . . . . . . . . . . . . . . 95
4.3 Riemannian Tucker manifold geometry . . . . . . . . . . . . . . . . . 97
4.3.1 Riemannian metric and quotient manifold structure . . 97
4.3.2 Characterization of the induced spaces . . . . . . . . . . . . 100
4.3.3 Linear projectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.3.4 Retraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.3.5 Vector transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.3.6 Computational cost . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.4 Algorithms for tensor learning problems . . . . . . . . . . . . . . . . 104
4.4.1 Tensor completion . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.4.2 General tensor learning . . . . . . . . . . . . . . . . . . . . . . . 106
4.5 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.5.1 Choice of metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.5.2 Low-rank tensor completion . . . . . . . . . . . . . . . . . . . 109
4.5.3 Low-rank tensor regression . . . . . . . . . . . . . . . . . . . . 113
4.5.4 Multilinear multitask learning . . . . . . . . . . . . . . . . . . 115
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
CHAPTER 5 Generalized thresholding for low-rank tensor
recovery: approaches based on model and
learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Fei Wen, Zhonghao Zhang, and Yipeng Liu
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.2 Tensor singular value thresholding . . . . . . . . . . . . . . . . . . . . . 123
5.2.1 Proximity operator and generalized thresholding . . . . 123
5.2.2 Tensor singular value decomposition . . . . . . . . . . . . . 126
5.2.3 Generalized matrix singular value thresholding . . . . . 128
5.2.4 Generalized tensor singular value thresholding . . . . . . 129
5.3 Thresholding based low-rank tensor recovery . . . . . . . . . . . . . 131
5.3.1 Thresholding algorithms for low-rank tensor recovery 132
5.3.2 Generalized thresholding algorithms for low-rank
tensor recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.4 Generalized thresholding algorithms with learning . . . . . . . . . 136
5.4.1 Deep unrolling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5.4.2 Deep plug-and-play . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.5 Numerical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
CHAPTER 6 Tensor principal component analysis . . . . . . . . . . . . . 153
Pan Zhou, Canyi Lu, and Zhouchen Lin
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
viii Contents

6.2 Notations and preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . 155

6.2.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.2.2 Discrete Fourier transform . . . . . . . . . . . . . . . . . . . . . 157
6.2.3 T-product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.3 Tensor PCA for Gaussian-noisy data . . . . . . . . . . . . . . . . . . . 161
6.3.1 Tensor rank and tensor nuclear norm . . . . . . . . . . . . . 161
6.3.2 Analysis of tensor PCA on Gaussian-noisy data . . . . . 165
6.3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
6.4 Tensor PCA for sparsely corrupted data . . . . . . . . . . . . . . . . . 166
6.4.1 Robust tensor PCA . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.4.2 Tensor low-rank representation . . . . . . . . . . . . . . . . . 172
6.4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.5 Tensor PCA for outlier-corrupted data . . . . . . . . . . . . . . . . . . 191
6.5.1 Outlier robust tensor PCA . . . . . . . . . . . . . . . . . . . . . 192
6.5.2 The fast OR-TPCA algorithm . . . . . . . . . . . . . . . . . . 196
6.5.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
6.5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
6.6 Other tensor PCA methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
6.7 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
CHAPTER 7 Tensors for deep learning theory . . . . . . . . . . . . . . . . . 215
Yoav Levine, Noam Wies, Or Sharir, Nadav Cohen, and
Amnon Shashua
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
7.2 Bounding a function’s expressivity via tensorization . . . . . . . . 217
7.2.1 A measure of capacity for modeling input
dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7.2.2 Bounding correlations with tensor matricization ranks 220
7.3 A case study: self-attention networks . . . . . . . . . . . . . . . . . . . 223
7.3.1 The self-attention mechanism . . . . . . . . . . . . . . . . . . 223
7.3.2 Self-attention architecture expressivity questions . . . . 227
7.3.3 Results on the operation of self-attention . . . . . . . . . . 230
7.3.4 Bounding the separation rank of self-attention . . . . . . 235
7.4 Convolutional and recurrent networks . . . . . . . . . . . . . . . . . . 242
7.4.1 The operation of convolutional and recurrent networks 243
7.4.2 Addressed architecture expressivity questions . . . . . . 243
7.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
CHAPTER 8 Tensor network algorithms for image classification 249
Cong Chen, Kim Batselier, and Ngai Wong
 Contents ix

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

8.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
8.2.1 Tensor basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
8.2.2 Tensor decompositions . . . . . . . . . . . . . . . . . . . . . . . 253
8.2.3 Support vector machines . . . . . . . . . . . . . . . . . . . . . . 256
8.2.4 Logistic regression . . . . . . . . . . . . . . . . . . . . . . . . . . 257
8.3 Tensorial extensions of support vector machine . . . . . . . . . . . 258
8.3.1 Supervised tensor learning . . . . . . . . . . . . . . . . . . . . . 258
8.3.2 Support tensor machines . . . . . . . . . . . . . . . . . . . . . . 260
8.3.3 Higher-rank support tensor machines . . . . . . . . . . . . . 263
8.3.4 Support Tucker machines . . . . . . . . . . . . . . . . . . . . . . 265
8.3.5 Support tensor train machines . . . . . . . . . . . . . . . . . . 269
8.3.6 Kernelized support tensor train machines . . . . . . . . . . 275
8.4 Tensorial extension of logistic regression . . . . . . . . . . . . . . . . 284
8.4.1 Rank-1 logistic regression . . . . . . . . . . . . . . . . . . . . . 285
8.4.2 Logistic tensor regression . . . . . . . . . . . . . . . . . . . . . 286
8.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
CHAPTER 9 High-performance tensor decompositions for
compressing and accelerating deep neural
networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Xiao-Yang Liu, Yiming Fang, Liuqing Yang, Zechu Li, and
Anwar Walid
9.1 Introduction and motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 294
9.2 Deep neural networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
9.2.1 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
9.2.2 Linear layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
9.2.3 Fully connected neural networks . . . . . . . . . . . . . . . . 298
9.2.4 Convolutional neural networks . . . . . . . . . . . . . . . . . . 300
9.2.5 Backpropagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
9.3 Tensor networks and their decompositions . . . . . . . . . . . . . . . 305
9.3.1 Tensor networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
9.3.2 CP tensor decomposition . . . . . . . . . . . . . . . . . . . . . . 308
9.3.3 Tucker decomposition . . . . . . . . . . . . . . . . . . . . . . . . 310
9.3.4 Hierarchical Tucker decomposition . . . . . . . . . . . . . . 313
9.3.5 Tensor train and tensor ring decomposition . . . . . . . . 315
9.3.6 Transform-based tensor decomposition . . . . . . . . . . . 318
9.4 Compressing deep neural networks . . . . . . . . . . . . . . . . . . . . 321
9.4.1 Compressing fully connected layers . . . . . . . . . . . . . . 321
9.4.2 Compressing the convolutional layer via CP
decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.4.3 Compressing the convolutional layer via Tucker
decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Visit https://ebookmass.com today to explore
a vast collection of ebooks across various
genres, available in popular formats like
PDF, EPUB, and MOBI, fully compatible with
all devices. Enjoy a seamless reading
experience and effortlessly download high-
quality materials in just a few simple steps.
Plus, don’t miss out on exciting offers that
let you access a wealth of knowledge at the
best prices!
x Contents

9.4.4
Compressing the convolutional layer via TT/TR
decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
9.4.5 Compressing neural networks via transform-based
decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
9.5 Experiments and future directions . . . . . . . . . . . . . . . . . . . . . 333
9.5.1 Performance evaluations using the MNIST dataset . . . 333
9.5.2 Performance evaluations using the CIFAR10 dataset . 336
9.5.3 Future research directions . . . . . . . . . . . . . . . . . . . . . 337
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
CHAPTER 10 Coupled tensor decompositions for data fusion . . . . 341
Christos Chatzichristos, Simon Van Eyndhoven,
Eleftherios Kofidis, and Sabine Van Huffel
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
10.2 What is data fusion? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
10.2.1 Context and definition . . . . . . . . . . . . . . . . . . . . . . . . 342
10.2.2 Challenges of data fusion . . . . . . . . . . . . . . . . . . . . . . 343
10.2.3 Types of fusion and data fusion strategies . . . . . . . . . . 347
10.3 Decompositions in data fusion . . . . . . . . . . . . . . . . . . . . . . . . 348
10.3.1 Matrix decompositions and statistical models . . . . . . . 350
10.3.2 Tensor decompositions . . . . . . . . . . . . . . . . . . . . . . . 351
10.3.3 Coupled tensor decompositions . . . . . . . . . . . . . . . . . 352
10.4 Applications of tensor-based data fusion . . . . . . . . . . . . . . . . 355
10.4.1 Biomedical applications . . . . . . . . . . . . . . . . . . . . . . . 355
10.4.2 Image fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
10.5 Fusion of EEG and fMRI: a case study . . . . . . . . . . . . . . . . . . 358
10.6 Data fusion demos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
10.6.1 SDF demo – approximate coupling . . . . . . . . . . . . . . 361
10.7 Conclusion and prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
CHAPTER 11 Tensor methods for low-level vision . . . . . . . . . . . . . . 371
Tatsuya Yokota, Cesar F. Caiafa, and Qibin Zhao
11.1 Low-level vision and signal reconstruction . . . . . . . . . . . . . . . 371
11.1.1 Observation models . . . . . . . . . . . . . . . . . . . . . . . . . . 372
11.1.2 Inverse problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
11.2 Methods using raw tensor structure . . . . . . . . . . . . . . . . . . . . 378
11.2.1 Penalty-based tensor reconstruction . . . . . . . . . . . . . . 379
11.2.2 Tensor decomposition and reconstruction . . . . . . . . . . 393
11.3 Methods using tensorization . . . . . . . . . . . . . . . . . . . . . . . . . . 409
11.3.1 Higher-order tensorization . . . . . . . . . . . . . . . . . . . . . 411
11.3.2 Delay embedding/Hankelization . . . . . . . . . . . . . . . . 413
11.4 Examples of low-level vision applications . . . . . . . . . . . . . . . 415
 Contents xi

11.4.1 Image inpainting with raw tensor structure . . . . . . . . . 415

11.4.2 Image inpainting using tensorization . . . . . . . . . . . . . 416
11.4.3 Denoising, deblurring, and superresolution . . . . . . . . 417
11.5 Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
CHAPTER 12 Tensors for neuroimaging . . . . . . . . . . . . . . . . . . . . . . . 427
Aybüke Erol and Borbála Hunyadi
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
12.2 Neuroimaging modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
12.3 Multidimensionality of the brain . . . . . . . . . . . . . . . . . . . . . . 431
12.4 Tensor decomposition structures . . . . . . . . . . . . . . . . . . . . . . 433
12.4.1 Product operations for tensors . . . . . . . . . . . . . . . . . . 434
12.4.2 Canonical polyadic decomposition . . . . . . . . . . . . . . . 435
12.4.3 Tucker decomposition . . . . . . . . . . . . . . . . . . . . . . . . 435
12.4.4 Block term decomposition . . . . . . . . . . . . . . . . . . . . . 437
12.5 Applications of tensors in neuroimaging . . . . . . . . . . . . . . . . 437
12.5.1 Filling in missing data . . . . . . . . . . . . . . . . . . . . . . . . 438
12.5.2 Denoising, artifact removal, and dimensionality
reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
12.5.3 Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
12.5.4 Registration and longitudinal analysis . . . . . . . . . . . . 445
12.5.5 Source separation . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
12.5.6 Activity recognition and source localization . . . . . . . . 451
12.5.7 Connectivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . 456
12.5.8 Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
12.5.9 Feature extraction and classification . . . . . . . . . . . . . . 463
12.5.10 Summary and practical considerations . . . . . . . . . . . . 468
12.6 Future challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
12.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
CHAPTER 13 Tensor representation for remote sensing images . 483
Yang Xu, Fei Ye, Bo Ren, Liangfu Lu, Xudong Cui,
Jocelyn Chanussot, and Zebin Wu
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
13.2 Optical remote sensing: HSI and MSI fusion . . . . . . . . . . . . . 488
13.2.1 Tensor notations and preliminaries . . . . . . . . . . . . . . . 488
13.2.2 Nonlocal patch tensor sparse representation for
HSI-MSI fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
13.2.3 High-order coupled tensor ring representation for
HSI-MSI fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
13.2.4 Joint tensor factorization for HSI-MSI fusion . . . . . . . 504
xii Contents

13.3 Polarimetric synthetic aperture radar: feature extraction . . . . . 517

13.3.1 Brief description of PolSAR data . . . . . . . . . . . . . . . . 518
13.3.2 The tensorial embedding framework . . . . . . . . . . . . . 519
13.3.3 Experiment and analysis . . . . . . . . . . . . . . . . . . . . . . 522
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
CHAPTER 14 Structured tensor train decomposition for speeding
up kernel-based learning . . . . . . . . . . . . . . . . . . . . . . . . 537
Yassine Zniyed, Ouafae Karmouda, Rémy Boyer,
Jérémie Boulanger, André L.F. de Almeida, and
Gérard Favier
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
14.2 Notations and algebraic background . . . . . . . . . . . . . . . . . . . . 540
14.3 Standard tensor decompositions . . . . . . . . . . . . . . . . . . . . . . . 541
14.3.1 Tucker decomposition . . . . . . . . . . . . . . . . . . . . . . . . 542
14.3.2 HOSVD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542
14.3.3 Tensor networks and TT decomposition . . . . . . . . . . . 543
14.4 Dimensionality reduction based on a train of low-order tensors 545
14.4.1 TD-train model: equivalence between a high-order TD
and a train of low-order TDs . . . . . . . . . . . . . . . . . . . 546
14.5 Tensor train algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548
14.5.1 Description of the TT-HSVD algorithm . . . . . . . . . . . 548
14.5.2 Comparison of the sequential and the hierarchical
schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
14.6 Kernel-based classification of high-order tensors . . . . . . . . . . 551
14.6.1 Formulation of SVMs . . . . . . . . . . . . . . . . . . . . . . . . 552
14.6.2 Polynomial and Euclidean tensor-based kernel . . . . . . 553
14.6.3 Kernel on a Grassmann manifold . . . . . . . . . . . . . . . . 553
14.6.4 The fast kernel subspace estimation based on tensor
train decomposition (FAKSETT) method . . . . . . . . . . 554
14.7 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
14.7.1 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555
14.7.2 Classification performance . . . . . . . . . . . . . . . . . . . . . 557
14.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
List of contributors

Kim Batselier
Delft Center for Systems and Control, Delft University of Technology, Delft,
The Netherlands
Yingyue Bi
School of Information and Communication Engineering, University of Electronic
Science and Technology of China (UESTC), Chengdu, China
Jérémie Boulanger
CRIStAL, Université de Lille, Villeneuve d’Ascq, France
Rémy Boyer
CRIStAL, Université de Lille, Villeneuve d’Ascq, France
Cesar F. Caiafa
Instituto Argentino de Radioastronomía – CCT La Plata, CONICET / CIC-PBA /
UNLP, Villa Elisa, Argentina
RIKEN Center for Advanced Intelligence Project, Tokyo, Japan
Jocelyn Chanussot
LJK, CNRS, Grenoble INP, Inria, Université Grenoble, Alpes, Grenoble, France
Christos Chatzichristos
KU Leuven, Department of Electrical Engineering (ESAT), STADIUS Center for
Dynamical Systems, Signal Processing and Data Analytics, Leuven, Belgium
Cong Chen
Department of Electrical and Electronic Engineering, The University of Hong
Kong, Pokfulam Road, Hong Kong
Nadav Cohen
School of Computer Science, Hebrew University of Jerusalem, Jerusalem, Israel
Xudong Cui
School of Mathematics, Tianjin University, Tianjin, China
André L.F. de Almeida
Department of Teleinformatics Engineering, Federal University of Fortaleza,
Fortaleza, Brazil
Aybüke Erol
Circuits and Systems, Department of Microelectronics, Delft University of
Technology, Delft, The Netherlands

xiii
xiv List of contributors

Yiming Fang
Department of Computer Science, Columbia University, New York, NY,
United States

Gérard Favier
Laboratoire I3S, Université Côte d’Azur, CNRS, Sophia Antipolis, France

Borbála Hunyadi
Circuits and Systems, Department of Microelectronics, Delft University of
Technology, Delft, The Netherlands

Pratik Jawanpuria
Microsoft, Hyderabad, India

Tai-Xiang Jiang
School of Economic Information Engineering, Southwestern University of
Finance and Economics, Chengdu, Sichuan, China

Paris A. Karakasis
School of Electrical and Computer Engineering, Technical University of Crete,
Chania, Greece

Ouafae Karmouda
CRIStAL, Université de Lille, Villeneuve d’Ascq, France

Hiroyuki Kasai
Waseda University, Tokyo, Japan

Eleftherios Kofidis
Dept. of Statistics and Insurance Science, University of Piraeus, Piraeus, Greece

Christos Kolomvakis
School of Electrical and Computer Engineering, Technical University of Crete,
Chania, Greece

Yoav Levine
School of Computer Science, Hebrew University of Jerusalem, Jerusalem, Israel

Zechu Li
Department of Computer Science, Columbia University, New York, NY,
United States

Athanasios P. Liavas
School of Electrical and Computer Engineering, Technical University of Crete,
Chania, Greece
 List of contributors xv

Zhouchen Lin
Key Lab. of Machine Perception, School of EECS, Peking University, Beijing,
China
Xiao-Yang Liu
Department of Computer Science and Engineering, Shanghai Jiao Tong
University, Shanghai, China
Department of Electrical Engineering, Columbia University, New York, NY,
United States
Yipeng Liu
School of Information and Communication Engineering, University of Electronic
Science and Technology of China (UESTC), Chengdu, China
Zhen Long
School of Information and Communication Engineering, University of Electronic
Science and Technology of China (UESTC), Chengdu, China
George Lourakis
Neurocom, S.A, Athens, Greece
Canyi Lu
Carnegie Mellon University, Pittsburgh, PA, United States
Liangfu Lu
School of Mathematics, Tianjin University, Tianjin, China
Yingcong Lu
School of Information and Communication Engineering, University of Electronic
Science and Technology of China (UESTC), Chengdu, China
George Lykoudis
Neurocom, S.A, Athens, Greece
Bamdev Mishra
Microsoft, Hyderabad, India
Michael K. Ng
Department of Mathematics, The University of Hong Kong, Pokfulam,
Hong Kong
Ioannis Marios Papagiannakos
School of Electrical and Computer Engineering, Technical University of Crete,
Chania, Greece
Bo Ren
Key Laboratory of Intelligent Perception and Image Understanding of Ministry of
Education of China, Xidian University, Xi’an, China
xvi List of contributors

Or Sharir
School of Computer Science, Hebrew University of Jerusalem, Jerusalem, Israel
Amnon Shashua
School of Computer Science, Hebrew University of Jerusalem, Jerusalem, Israel
Ioanna Siaminou
School of Electrical and Computer Engineering, Technical University of Crete,
Chania, Greece
Christos Tsalidis
Neurocom, S.A, Athens, Greece
Simon Van Eyndhoven
KU Leuven, Department of Electrical Engineering (ESAT), STADIUS Center for
Dynamical Systems, Signal Processing and Data Analytics, Leuven, Belgium
icometrix, Leuven, Belgium
Sabine Van Huffel
KU Leuven, Department of Electrical Engineering (ESAT), STADIUS Center for
Dynamical Systems, Signal Processing and Data Analytics, Leuven, Belgium
Anwar Walid
Nokia Bell Labs, Murray Hill, NJ, United States
Fei Wen
Department of Electronic Engineering, Shanghai Jiao Tong University, Shanghai,
China
Noam Wies
School of Computer Science, Hebrew University of Jerusalem, Jerusalem, Israel
Ngai Wong
Department of Electrical and Electronic Engineering, The University of Hong
Kong, Pokfulam Road, Hong Kong
Zebin Wu
School of Computer Science and Engineering, Nanjing University of Science
and Technology, Nanjing, China
Yang Xu
School of Computer Science and Engineering, Nanjing University of Science
and Technology, Nanjing, China
Liuqing Yang
Department of Computer Science, Columbia University, New York, NY,
United States
 List of contributors xvii

Fei Ye
School of Computer Science and Engineering, Nanjing University of Science
and Technology, Nanjing, China
Tatsuya Yokota
Nagoya Institute of Technology, Aichi, Japan
RIKEN Center for Advanced Intelligence Project, Tokyo, Japan
Zhonghao Zhang
School of Information and Communication Engineering, University of Electronic
Science and Technology of China (UESTC), Chengdu, China
Qibin Zhao
RIKEN Center for Advanced Intelligence Project, Tokyo, Japan
Guangdong University of Technology, Guangzhou, China
Xi-Le Zhao
School of Mathematical Sciences/Research Center for Image and Vision
Computing, University of Electronic Science and Technology of China, Chengdu,
Sichuan, China
Pan Zhou
SEA AI Lab, Singapore, Singapore
Ce Zhu
School of Information and Communication Engineering, University of Electronic
Science and Technology of China (UESTC), Chengdu, China
Yassine Zniyed
Université de Toulon, Aix-Marseille Université, CNRS, LIS, Toulon, France
This page intentionally left blank
Preface

This book provides an overview of tensors for data processing, covering computing
theories, processing methods, and engineering applications. The tensor extensions
of a series of classical multidimensional data processing techniques are discussed
in this book. Many thanks go to all the contributors. Students can read this book to
get an overall understanding, researchers can update their knowledge on the recent
research advances in the field, and engineers can refer to implementations on various
applications.
The first chapter is an introduction to tensor decomposition. In the following, the
book provides variants of tensor decompositions with their efficient and effective so-
lutions, including some parallel algorithms, Riemannian algorithms, and generalized
thresholding algorithms. Some tensor-based machine learning methods are summa-
rized in detail, including tensor completion, tensor principal component analysis,
support tensor machine, tensor-based kernel learning, tensor-based deep learning, etc.
To demonstrate that tensors can effectively and systematically enhance performance
in practical engineering problems, this book gives implemental details of many ap-
plications, such as signal recovery, recommender systems, climate forecasting, image
clustering, image classification, network compression, data fusion, image enhance-
ment, neuroimaging, and remote sensing.
I sincerely hope this book can serve to introduce tensors to more data scientists
and engineers. As a natural representation of multidimensional data, tensors can be
used to substantially avoid the information loss in matrix representations of multiway
data, and tensor operators can model more connections than their matrix counterparts.
The related advances in applied mathematics allow us to move from matrices to ten-
sors for data processing. This book is promising to motivate novel tensor theories and
new data processing methods, and to stimulate the development of a wide range of
practical applications.

Yipeng Liu
Chengdu, China
Aug. 10, 2021

xix
Visit https://ebookmass.com today to explore
a vast collection of ebooks across various
genres, available in popular formats like
PDF, EPUB, and MOBI, fully compatible with
all devices. Enjoy a seamless reading
experience and effortlessly download high-
quality materials in just a few simple steps.
Plus, don’t miss out on exciting offers that
let you access a wealth of knowledge at the
best prices!
Other documents randomly have
different content
 you see a driver with a yellow plaid cap. We are
relying on you not to have anybody with you, or in
watching,—if you do, we shall know it, and the
whole deal is off. You will not hear from us again. If
you attempt anything,—anything at all but the most
perfect good faith and honesty in your course, you
will be more than sorry. In a word, you will then
bring about the sudden death of the man you love.
There is no more to be said on that score. Get into
the taxi and when it stops, near another taxi, make
a quick change. Have the money with you in a
small compact parcel. The second taxi will take you
along a certain road. When it meets a certain car, it
will slow down and you will hand the parcel to the
man who leans out of that car for it. That is all.
Good-bye.

Elsie read and re-read the missive.

She was uncertain what to do. Her impulse was to lay 208
the whole matter before Whiting or Coleman Coe, and
follow their advice.

But suppose they should say,—as so many people do,—

make no bargains with the kidnappers. Treat any such
communications with silent contempt,—or, arrange for
police protection, even if it is forbidden.

The more she thought it over, the more she was inclined
to manage the whole affair alone. She could do it,—and
she was not afraid. It was all to be done in broad
daylight, there was no danger if she herself acted in
good faith. And if she brought any one else into it, there
was grave danger, not only to herself but to Kimball.
She looked curiously at the card that had come in the
letter.

It was an address on Broadway, and was evidently,— 209

even to her inexperienced mind,—the office of a loan
broker.

From him she could get the necessary money on the

assurance of her nearby wedding and consequent
inheritance. Arrangements had, of course, been made
by the perpetrators of the crime against Kimball Webb.
They must be a clever and powerful set,—they were so
unafraid of anything or anybody. The thought of her
restored lover and their wedding at last, so thrilled Elsie,
that she began preparations at once.

She could scarcely control her impatience to get to the

broker’s office.

Once there, she found indeed, that all had been

arranged.

The affable Hebrew, who presided over the

establishment, was confidentially minded, and was quite
ready to advance the large sum required in return for
Elsie’s signed promise to pay,—with exorbitant interest,
the day after her marriage.

For Elsie Powell and her affairs were well known to

newspaper readers and the affable Jew felt no qualms
of doubt as to his future reimbursement and his usury.

The parcel, made up neatly and inconspicuously, was

handed to Elsie and her signed document carefully put
away in a big safe.
The transaction meant little to Elsie, herself, so wrapped 210
up was her whole soul in her coming adventure.

She would get Kimball back! That was all she knew or
cared about!

She went to Altman’s, her precious package in her

handbag, which she carried with seeming carelessness,
but with a watchful eye.

She had a strange feeling of security because of the

character and appearance of the notes she had
received. Had they been illiterate scrawls she would
have hesitated to go ahead as she had done, but the
educated and socially correct tone of the letters gave
her the impression of brains and character, however big
a villain the writer might be.

With a beating heart, but with a steady step she came

out of Altman’s shop and seemed to glance casually
about for a cab.

Seeing a driver with a yellow plaid cap, she beckoned

him and got into his cab.

No word was spoken as she settled herself on the seat,

and watched the man start the car.

He, too, was nonchalant of manner, and drove away

toward Madison Avenue.

From there they followed a devious course, turning

often, returning on their own tracks, wheeling suddenly,
performing various eccentric detours, all, doubtless in
an endeavour to detect a follower, if any.
Elsie sat quietly, unmoved by these strange motions, 211
and full of buoyant hope that all would be well, since
she had not betrayed her trust.

After a time the taxicab stopped at a curb, another cab

drew up at its side, and Elsie stepped from one to the
other.

The second cab had also a taciturn, grave-faced driver.

Though he said no word, gave no look of intelligence,
Elsie felt a sense of safety with him, from his very
silence. She was free from all fear, and looked forward
eagerly to the consummation of her errand.

This time it was a long drive. On they went, northward

from the city and into a pleasant, wooded locality.
Swiftly the car flew and after an hour’s journey they
were on a smooth road, with groves of trees on either
side. But it was a travelled road, and its well-kept
asphalt proclaimed its nearness to civilization.

Elsie kept her eyes open and her mind clear. She grew
impatient for the end of her trip, but she preserved her
poise and her balance.

“Here’s the car, miss,” the taxi driver said suddenly, and
she saw a red roadster approaching swiftly.

Both cars slowed down and then stopped.

From the red car a man leaned out. He had a small 212
mask on that concealed most of his features, but Elsie
caught a gleam of many gold filled teeth in his lower
jaw. Into his outstretched hand, conveniently near, Elsie
placed the packet, from her hand-bag. She felt a shock
of disappointment that she did not receive Kimball in
return, right then and there, but she had no time to
speak. In a flash, the driver on the cab she was in,
sprang from his seat, jumped into the red car, and like a
streak the roadster disappeared.

Alone, in a driverless taxicab, Elsie sat, unable for a

moment to realize what had happened.

Slowly it dawned upon her that she had been tricked,—

swindled,—but no, she couldn’t believe that! She felt
sure that the men had only carried out their plans for
safety. That they feared pursuit and had made off with
the money and would restore Kimball in their own good
time, she had no doubt. The thing was, now, how was
she to get home?

She wasn’t greatly alarmed, for the well-kept road gave

hope of frequent travellers, and somebody would take
her back to New York.

And, after a time, somebody did. She let several cars

pass before she asked help, and though curious looks
were cast at her, no one intruded upon her. But when
she saw a car come by, with a good chauffeur, and a
benignant looking lady in the tonneau, she asked for a
ride to New York.

The benignant looking lady was not all that could be 213
hoped for in the way of cordiality, but when Elsie
explained that the taxicab had refused to go and the
chauffeur had gone for help and that she was in great
haste to get to the city the lady agreed to take her.
Remarking, however, that for a girl who wanted to get
to New York in haste, her cab was turned astonishingly
in the opposite direction!

But Elsie’s smile and winning manner soon overcame

the other’s asperity, and they were affably chatting long
before they reached the city.

Naturally enough, the kind lady asked the name of her

passenger, but Elsie, knowing the necessity for caution,
gave an assumed name and address and made up a
story of her life that was as plausible as it was false.

But she dared take no chances on breaking her pledge

of inviolate secrecy, lest she lose her chance of getting
Kimball back, and after all she had gone through, that
would be unbearable.

She asked to be set down at the Grand Central Station,

as she was going back to her home,—avowedly in
Boston,—that night.

Warmly friendly by this time, the benignant lady set her

down as requested, after exacting a promise to hear
from her by letter.

Alone again, Elsie flew for a taxicab and went straight 214
home. She glanced at the mail, arrived since her
departure, but was not surprised to find no letter in the
writing of her new correspondent. He had said he would
not write again, and she did not think he would.

She had nothing to do now, but wait. She had

conscientiously fulfilled her part of the bargain, and she
had utter faith that the abductors of Kimball would do
the same. They had their money—what more did they
want?

She waited all that evening, dully patient, quietly serene

of manner, but with a heart that beat wildly when the
door bell or telephone sounded.
Occasionally, she telephoned to the Webb house, hardly
thinking Kimball would go there before coming to her,
but unable to resist general inquiry.

At bedtime, she had heard nothing from him, and

resolved to go to bed and to sleep in happy hopes of a
blessed meeting tomorrow.

She could not sleep,—slumber does not come for the

willing of it and as she tossed in wide awake suspense,
her thoughts took a new turn.

Suppose,—just suppose she had been tricked! Suppose 215

the notes had not come from the men who stole
Kimball,—ah, they must have done so! She had Kim’s
note to prove it! Nothing ever could make her believe
that note a forgery. She knew his dear writing too well—
she knew every stroke of his pen, every peculiarity of
his really unusual handwriting, and she felt in every
letter of that note that he himself had penned it. There
was no chance that he had not. Therefore, the letters
from the kidnappers were in good faith. They proved
the fact that Kimball had been abducted,—and held for
ransom. Well, now they had the ransom, and Kim would
be returned. Of course he would! She would not think
otherwise, or she would die! She knew he would come
tomorrow,—and in that knowledge she at last fell
asleep.

She awoke with a start. Throwing on her night light, she

found it was three o’clock in the morning. She felt a
strange numbness of mind, a peculiar feeling as if the
end of the world had come. Striving to determine what
it all meant, she realized that she had lost hope,—that
she was now persuaded that she had been tricked. The
notes were from the kidnappers but they had no
intention of returning her lover!

Something, she could not tell what, brought the

conviction to her soul that she had done very wrong in
following their bidding blindly in giving them the money
on such uncertainty. She remembered clearly the smile
of the man in the red car,—the smile that had disclosed
those gold-filled teeth, and she knew she had been
duped, deceived and swindled!

216
 CHAPTER XIV
AN EASY MARK

Though slow to anger, Elsie was a little firebrand when

roused. And the more she thought over the matter the
more furious she grew at the game that had been
played on her. The fact that she brought it all upon
herself only made her more angry.

And, yet, she didn’t blame herself utterly, for she had
felt so sure that only by following instructions implicitly,
could she accomplish her end.

She didn’t for a moment believe that some one had

tricked her who knew nothing of Kimball Webb, for she
had his own letter to disprove that. She concluded they
had tricked him, too, and had forced him to write the
note and then had cheated him as they had her.

Still, he might come home yet; the day might bring him
or news of him.

But when the slow hours passed and morning melted

into afternoon, poor Elsie gave up hope.

By the time Coe came in the evening, Elsie had decided 217
to tell him the whole story, assuming that since the
money was paid, it was now no breach of trust.
Coley Coe stared at her as she unfolded the surprising
tale.

“You chump! You Easy Mark!” he cried, angrily, quite

forgetting in his astonishment to whom he was
speaking.

“I beg your pardon,” he said, as he noted her rising

colour. “I oughtn’t to say such things,—but, oh, Miss
Powell, how could you go off on such a wild-goose
chase,—and a dangerous one, too?”

His thatch of hair bobbed wildly about in his excitement,

and he clutched at it as if almost frenzied.

Then he calmed down, and looked at the thing squarely.

His blue eyes seemed to grow darker as their
concentrated gaze fell on Elsie’s troubled face.

“It’s outrageous!” he cried, “it’s a shame, but, Miss

Powell, the villains may have overreached themselves.
They may have started something that will lead to their
own undoing. We’ve learned a heap from this
experience of yours. Now, tell me all over again,—every
smallest detail.”

So again Elsie went over the whole story, and told of

every step of the way.

“Clever! clever!” was Coe’s grudging tribute to the ability

of the abductors.

“You see the first taxicab was a real one. They engaged 218
the driver to do just what he did do. The second was a
fake one,—their own car and one of their own men.
Then when the time came, the car was abandoned,—
and so were you. They knew you’d get a lift back to the
city,—and they didn’t care whether you did or not! In
one way, I can’t blame you, Miss Powell, for I see you
didn’t dare tell me. Yet, you might have known they’d
not release their prisoner.”

“I don’t agree,” cried Elsie. “How could I know that? And

if they had given him to me the money was well spent.”

“That’s so; it wouldn’t have been surprising if they had

let him go; they’d doubtless be glad to get rid of him.
But I think your quick willingness to give the money
make them greedy for more, and I think they’ll try the
same game right over again.”

“Oh,” Elsie cried, “I couldn’t do it again!”

“No, indeed! And you’re not going to throw away

another fifty thousand dollars, if I can prevent it! Now,
let’s consider. What have we learned? What sleeping
dogs have we stirred up? Much depends on the positive
fact that this note is really from Mr. Webb himself.
You’re sure?”

“Absolutely,” declared Elsie. “I know Kimball’s writing, 219

and I know that’s it. Nobody could forge so skilfully,—
you can see that yourself. It’s dashed off.”

“Yes, that’s so. A forgery would show a little hesitation

or painstaking effort. But I’m going to show it to an
expert. He can tell if he has some of Webb’s other
letters.”

“Anybody could tell,” insisted Elsie. “Wait, I’ll get some

letters.”

She ran away to her own room and returned with a

packet of them.
Comparison soon made it evident that the note in
question was beyond all doubt the work of Webb
himself. A thousand little points proved it. Coe was
satisfied, and went on with his conclusions from it.

“You see, it proves a whole lot of things,” he cried,

jubilantly. “Perhaps your money, enormous sum though
it was, bought worthwhile evidence.”

“Such as what?”

“Well, to begin with, we know now that Webb was really

abducted, and is now held against his will. This does
away with all thought of his having decamped on
purpose,—also, to my mind, precludes the theory of his
mother or sister being implicated. Miss Webb is a Tartar,
—if you ask me! but she never managed the affair of
yesterday!”

“No, she never did! Henrietta is not acquainted with

those—”

“Loan Sharks! Right! Kimball Webb was carried off by 220

desperate and clever men,—and, here’s a strong point,
—he was unconscious when removed from his room.”

“How do you know?”

“Because in this first letter, it says the means used will

never be known by any one,—not even himself. So, as I
imagined, he was taken from his room,—from his home,
while unconscious,—in a drugged sleep probably, and
therefore, we must assume a secret entrance!”

“But there isn’t any!”

“There is! There’s got to be! They couldn’t take him
through the door and fasten it behind them! They
couldn’t get him out of that six inch opening at the top
of a window! There has to be a secret way out! And, by
George, I’m going to find it, if I have to tear the house
down!”

“I’d rather you’d find Kim,” said Elsie, sadly.

“You poor child! Of course you would. Forgive me, I’m

afraid I seem to think less of the quarry than the chase!
But I don’t really. We’re going to get Kimball Webb
back,—and we’re going to do it by means of the
information you unconsciously achieved through this
adventure of yours!”

“And you don’t think they mean to give him back after I
did my part?”

“I do not! They look on you as an inexhaustible gold 221

mine. They’ll wait a while and then make a stab for
another big sum. Less maybe than the first, but
exorbitant. Apparently they’re not afraid of anything or
anybody. Clever chaps, but sure to come a cropper yet!”

“How do you know?”

“Oh, they’re too cocksure; they’re bound to overlook or

forget some little thing, and now I know there is a scent
to be followed, I’m all for following it. Now I know
there’s a sleeping dog, I shan’t let him lie! Take that
letter! The two letters from them! Look at ’em! No
attempt at disguised writing. Plain, bold penmanship,—
not printed nor words cut out from a newspaper, nor
any of those hackneyed stunts.”

“Well?”
“Well, that proves they were written by some one who
never could by the remotest chance be suspected.
Somebody so outside suspicion that they’re willing to
send his regular handwriting.”

“Proving?”

“Proving a clever, bold master spirit, who stops at

nothing and who knows just what he dare do and what
not! I believe he fully intended to set Mr. Webb free on
the receipt of the money,—then, when you proved such
a ninny,—pardon me, it slipped out,—but you were!
then, he concluded you were good for one more touch,
at least.”

“Well, if what I learned,—or made it possible for you to 222

learn—restores Kimball Webb to me,—I’ll never
begrudge the money.”

“That is, if we get him home in time for the wedding.”

“Oh, I don’t care for the fortune—”

“Then, just how are you going to pay your indebtedness

to the Hebrew gentleman?”

Elsie’s face fell. “I hadn’t thought of that!”

“It’s a big thing to think of, Miss Powell! You can’t get
out of that obligation, you know. And while the receipt
of your aunt’s money would make it easy for you to pay
it, yet if you are not married by your birthday—”

“And do you think if I had acted differently in any way, I

could have held those men to their agreement?”

“I can’t say positively,—but I do think so.”

“What ought I to have done?”

“Demanded the person of Webb before you gave up the

money,—or at least, asked for some assurance of his
return, and asked when and where you might expect to
see him.”

“I was too frightened.”

“I know you were, and they knew it, too.”

“And anyway, even if they had made me promises then,

they wouldn’t have kept them.”

“Likely not. Now, Miss Powell, here’s a hard fact,—if Mr.

Webb is not here by your birthday, you’ll have to marry
somebody,—in order to get that money so you can pay
off that loan.”

“What?” Elsie’s face went white, and her eyes were 223
filled with horror at the sudden realization of the truth
of Coe’s statement.

“I won’t,—I’ll kill myself first!”

“Oh, come now, don’t talk about killing. And that would
be a cowardly thing, for your people would be hounded,
—whether legally or not.”

“Mother and Gerty! Oh, no!”

“I don’t say they could be made to pay it, but there’d be

some mighty unpleasant experiences coming to them!
No, Miss Powell, don’t kill yourself,—surely a marriage
with some man other than Mr. Webb would be a better
fate than suicide!”
“No, not to my way of thinking. But I must think of my
mother and sister! Oh, Mr. Coe, do help me! I think I
shall go distracted!”

“Small wonder! You poor child! I wish, now, we had

more time. The birthday is drawing perilously near.
Something must be done. Of course, you can’t describe
either man well enough for positive identification?”

“No; the taxi driver, the second one was a decent 224
looking man, of medium build, with a grave, rather
stern face. He was dark, I think,—with brownish hair. I
saw his back mostly, and didn’t notice his face at all. I
thought of him merely as a means to an end, and when
the red car came along, I thought only of giving up the
money. And the man in the red car wore a mask,—just
a small one, but it covered his eyes and nose and came
down partly over his mouth. But I noticed several gold
filled teeth in the lower jaw. Unusually bright they
were.”

“That would be a help, if we could get any other hint

which way to look. But, as I said, the master mind
behind all this scheme is so diabolically clever, that he
has discounted all chances of discovery and, I’ve no
doubt, feels secure from police and detectives.

“Now, I’m for spending another night in that room of

Kimball Webb’s, and I’ll bet there’ll be no Poltergeist this
time!”

“Why?”

“Why, don’t you see it! The arch villain,—I feel sure
there’s one principal and two or more subordinates,—
the chief devil, we’ll say, has a means of access to that
room. It was he who was responsible for all the
Poltergeist performances, he who pulled bedclothes off
Webb, and later, off yours truly,—he who made a ghost
appear,—”

“How?”

“Oh, lots of ways for that. I’ll tell you some other time. I 225
must skittle, now. Go to sleep and dream of Webb’s
return. But,—and this is very serious, Miss Powell,—if I
don’t succeed in getting him back,—if the villains are
scared off or any such matter, you must make up your
mind to marry somebody else. For I should hate to see
you in the clutches of that wretch of a Loan Broker!
You’ve no idea what it would mean!”

Coe went away, and Elsie went straight to her room.

She denied admittance, when Gerty begged for it, and
said she wanted to rest.

But rest, she did not; in fact she was such a victim of
unrest, worry and anguish, that morning found her in a
high fever and grave danger of nervous collapse.

The doctor came, a nurse was summoned and for a few

days brain fever was feared. But Elsie’s strong
constitution and brave will power conquered, and she
pulled through without the dreaded attack.

The doctor ordered, however, a change of scene, were

it ever so small a journey, and after some discussion
Elsie agreed to go to Atlantic City for a few days.

Coley Coe was the one who finally persuaded her to

adopt the plan. He promised to keep in constant touch
with her and tell her any bit of information he could
gain. He said he would come down to see her as often
as necessary for their mutual conference, and he felt
sure that she would be better off in every way from her
family for a time.

He had slept in Kimball Webb’s room several nights,

since, and as he anticipated, nothing at all had
happened.

“You see,” he said, “the rascal thought he could make it 226

appear supernatural, now he knows I’m on his trail, he
has given up that idea.”

“How does he know it?” asked Elsie. “Is he omniscient?”

“Nearly so! You may depend he knows every step that is

taken toward his discovery! Why, Miss Powell, he’s a
man in the know, every way. He may not be one of Mr.
Webb’s own particular circle, socially, but he’s enough in
his set or in his life somehow, to be in touch with
everybody even remotely connected with the case.”

“Have the police done nothing at all?”

“Yes, they’re working at it. But their methods are

different from mine, and while they’re all right, I doubt
if they get anywhere. Sometimes I doubt if I will, either.
Howsumever, you toddle along to Atlantic City with
Nursey, and I’ll try to corral a nice young man for you to
marry before the fatal thirtieth gets much nearer. You
wasted some good time with that illness of yours,—
though I don’t wonder at it, I’m sure.”

“Why, what could I have done,—if I hadn’t been ill?”

“Nothing definite, but I feel sure the abductors would 227

have written you another of those good-looking notes,
and if you had gone on another taxi ride, I should have
been off in the offing somehow.”
The nurse, a Miss Loring, was a pleasant, sympathetic
girl, and as she of course knew all about Elsie’s tragedy
from the papers, she was deeply interested in her young
charge. She was experienced and capable and Elsie
found herself really glad to go away with the kind and
gentle nurse.

They were pleasantly located in The Turrets, a new

hotel, and after twenty-four hours of rest and sea air
Elsie felt wonderfully better.

“I’m not really ill, you know,” she said, and the nurse
agreed.

“No, Miss Powell, but it was a real nervous breakdown,

and another will follow, unless you try to keep it off.”

“I’ll try,” and Elsie voluntarily became a biddable and

obedient patient.

It was on a Thursday,—just one week before the

thirtieth of June that the two went for a ride in the
rolling chairs. Sometimes they rode together, but this
day they chanced to take separate chairs.

The man who pushed Elsie’s was a big, husky chap,

with an engaging smile. Miss Loring’s man was a slender
youth, but of a wiry strength.

For a time they rode close together, chatting casually,

and then as Elsie grew silent, the nurse ceased to
bother her with talk.

Thus, it chanced, now and then, one chair or the other 228
forged ahead, by reason of the traffic or danger of a
collision.
And one time, when Elsie’s chair was pushed ahead of
Miss Loring’s it did not fall back beside the nurse’s chair
as promptly as usual.

Elsie looked around for the nurse, but failed to see her.

“Where’s my companion?” she said over her shoulder;

“don’t let us get separated.”

“No, ma’am,” smiled the big man who pushed her, and
she settled back into her seat, thinking deeply.

A moment later, she looked around again, and still not

seeing the nurse told the man to wait for her to come
up to them.

“Why, the other lady is ahead, ma’am, I’ll catch up to

her,” and he moved her chair more quickly.

Elsie looked about with a sudden thrill of alarm, and

saw no sign of the nurse anywhere.

“Here we are, ma’am, she just went in here,” the man

stopped the chair in front of a tall hotel.

“Went in here? What do you mean?”

“Yes’m, the lady who belongs with you,—the nurse,

ma’am, she went in here in great haste and motioned
for you to follow her. Better go in, ma’am.”

Bewildered, Elsie allowed herself to be assisted from the 229

chair and ushered inside, not thinking at the moment
that it was strange for the chair-pusher to be so
officious.
“What in the world did Miss Loring come in here for?”
she asked, as they stood a moment in the hall.

“I don’t know, ma’am, but I just saw her go up in this

elevator. She beckoned for you to follow.”

Elsie hesitated a moment, but it was a first class hotel,

not a large building but a tall one, and handsomely
appointed.

She got into the elevator, the man following, indeed,

urging her in by a guiding hand on her elbow.

“Tenth,” he said to the elevator girl, and the car shot

upward.

It was not until they were walking along the corridor on

the tenth floor that Elsie felt a thrill of fear. What did it
mean? Surely Miss Loring never came up here,—
expecting Elsie to follow!

“Here you are,” and as they reached a closed door, the

man swung it open and led Elsie firmly inside. “Sorry,
Miss, but I’m only obeying orders. Good-bye.” He jerked
off his cap, closed the door behind him and went away,
leaving Elsie alone, in a strange room in a strange
house.

She flew to the door, but she could not open it. She was 230
trapped,—and she had walked into a trap, unresistingly,
in broad daylight!

What would Coley Coe say to her now?

She went to the window and looked out. The familiar

sight of the ocean and the boardwalk cheered her. She
didn’t know what she was to experience next, but she
felt a sense of relief at sight of the throngs of people.

She was alone in the room for what seemed hours but
was not more than twenty minutes when the door was
flung open and in rushed,—not the man with the gold
teeth, whom she had rather expected to see,—but Fenn
Whiting.

“Oh, Elsie,” he cried, wildly, “am I in time?”

“Time for what?” she asked bewilderedly.

“Why, I met Miss Loring and she said she had lost you,
and I chased madly about asking everybody questions,
and I finally traced you here! Who brought you? What
does it mean?”

“I know no more than you do, Fenn,” and so relieved at

sight of a kind and familiar face was she, that Elsie
burst into tears on his shoulder.

“There, there, darling,” he soothed her, “never mind,—

it’s all right. Stay there, dearest, that’s your rightful
place. I hope it will always be your haven in troublous
times. Be quiet, my love, don’t try to talk yet,—and
when you can, then tell me what happened.”

“Yes, I can talk! I’m all right,” and Elsie stopped crying; 231
“I’m only mad! Why, Fenn, somebody trapped me into
this room!”

“Trapped you! What do you mean?”

“Just that!” and Elsie told how the chair-pusher had led
her to the house, and urged her up in the elevator and
into the room, and then had locked her in.
“Why, the door isn’t locked,” Whiting exclaimed, “I
walked right in!”

“How did you know I was in here?”

“Asked the elevator girl,—she told me.”

“Well, the door was locked on this side,—must be a

spring catch.”

“It must be, then,”—and Whiting went to examine it.

“Yes, it is. Thank Heaven I could open it from outside.
Well, dearest, we’ll go home, shall we?”

“Yes, I suppose so. But I want to know what it all

means.”

“Didn’t you know your chair man?”

“No; we pick up different ones every time,—wherever

we happen to be. He wasn’t a real one, of course. He
must have been placed there, so I’d engage him, by
those villains—”

“What villains? What are you talking about?”

Elsie bit her lip. She had promised Coe to reveal no 232
slightest word regarding her experiences with the
kidnappers of Webb, and now she had given a hint!

“Nothing,” she said, “nothing, Fenn. Oh, I am ill, please

take me home!”

“You’re not ill, Elsie, but you’re terribly frightened. Tell

me what about and tell me who are the villains who are
troubling you. Let me settle with them! I am your
rightful protector. You are engaged to me, and in less
than a week is our wedding day! Can’t we announce it,
at once, and let me be known as your proper protector?
You shall not leave this room until you say yes!”

233
 CHAPTER XV
IN UNIFORM

“Is that a threat?” Elsie turned on Whiting, with sudden

rage.

“Not unless you choose to take it so.” But the man’s

steely grey eyes were commanding rather than
imploring, and his thin lips were set in a straight line
that bespoke determination. “Don’t make me threaten
you, Elsie,—why should it be necessary? I love you and
I want you,—but more than that I want your promise to
marry me at once to save yourself from persecution and
trouble. You were trapped here, you say,—you just
referred to some villains who have, I must infer, already
annoyed you. Why haven’t you told me of it?”

“Why should I? I can’t marry you, Fenn, after all. I know

I said I would,—and you know what I said I’d do right
afterward. But I can’t do that. Perhaps I’m too much of
a coward, to take my own life,—perhaps it would be a
cowardly thing to do, anyway. But, I can’t marry you—”

“You must, Elsie, you promised me—”

“Such promises have been broken before this! A consent 234

to marry is not a marriage contract! Sue me for breach
of promise, if you choose,—I refuse to marry you!”
Her voice rose at the last to an almost hysterical shriek.
She was both nervous and frightened. The knowledge
that she had been abducted,—for that was what it
seemed to be,—scared her, and though grateful for
Whiting’s rescue and his presence, yet she felt a strange
fear of him, too.

“Let me go,” she said, at last, starting toward the door.

“No,” and Fenn strode across the room, locked the door
and pocketed the key. “No, you shall not go until I have
your promise,—and an unbreakable one this time. In
fact, Elsie, I want you to marry me right now and here.
I’ll arrange all details,—I have arranged most of them.
Just consent, dearest, and then you’ll be mine to love
and care for and to protect from those villains you
speak of.”

“Fenn, are you crazy?”

“No, I’m not, but you’ll be, if you keep up this nervous
tension you’re living under. Be guided by me, Elsie,
darling; marry me out of hand, and we’ll go away to
some beautiful, quiet spot, and all care shall be lifted
from your dear shoulders.”

Elsie looked at him curiously.

“Suppose I agree to marry you the day after my

birthday,” she said; “will that do?”

“Do perfectly, as far as the loss of your fortune is 235

concerned. I’ve told you before I’m no fortune hunter.
You must believe it by now. I’d rather marry you at
once, for your sake, and for my own. But not for the
sake of the inheritance. So, promise me sacredly to
marry me the day after your birthday, and I’ll take you
home now.”

“Oh, no, Fenn, don’t you see, if I marry you, it must be

before the thirtieth, to get the money for Mother and
Gerty. They’d never forgive me otherwise. And, too, why
should I wait? I’d like the money all right,—if only I
didn’t have to marry to get it. What an awful will! And
yet, it all seemed so lovely when I had Kimball with
me!”

“It will seem just as lovely when I’m with you. Let me
try, dear; give me a chance to make good! I’m not over
conceited, but I’m sure I can make you happy. If you
choose to marry me in time to get the money, we can
do wonderful things! Take wonderful trips, see beautiful
places,—but beautiful to me, only because you are with
me!”

There was a deep thrill in his tones that moved Elsie by 236
its genuine passion and devotion. She looked into his
grey eyes,—their steely glint softened now, and read
there a great unconquerable love for herself. Should she
cast this aside for a chance, an uncertainty? She must
get the money for her people,—she had decided on
that,—and she felt it her duty to sacrifice herself for
them. But, when she tried to say yes to Whiting’s pleas,
the word would not come.

“I can’t! Oh, Fenn, I can’t!” she moaned. “I love Kimball,

—oh, I love him desperately! I can never marry any one
else,—I can’t—I can’t do it!”

“Hush, Elsie, don’t sob so. Listen, dear; the time for that
sort of thing is past. There are only seven days now to
your birthday; you can’t wait till the last minute to
decide. And if Webb had been coming back he would
have been here before this. He will never come back,—
I’m sure of it!”

“You can’t be sure of it, Fenn; but will you arrange it

this way,—you said you would, once. Let the wedding
take place the day before my birthday, and if Kim comes
home, let him be the bridegroom, and if not, I’ll marry
you.”

“No! I’ll not do that! You’ve played fast and loose with
me long enough! I’ve stood for it because I love you so,
and I want you so. But I won’t be that sort of a cat’s-
paw! You’ll say right now you’ll marry me, or I’ll drop
out of it all, and you can marry anybody you choose to
get your precious legacy!”

Whiting’s face was distorted by passion and by rage at 237

the idea of being baffled at the last. “I do not think for a
minute that Webb would show up, but if he did, I’d not
stand having my bride snatched from me at the very
altar! No!”

“Then, you may drop out!” Elsie’s determination was as

great as his own. “I refuse to promise. I’d rather marry
Joe Allison, at the last minute, and so keep a chance for
Kim, than to promise you, and have no chance at all!”

“Allison! You would, would you? We’ll see about that!”

Whiting quite lost control of himself and flew into a

veritable frenzy. “You’ll marry me now, and here,—get
that?”

Elsie was horror-stricken. Fenn’s teeth were set together

and his expression was that of a hungry, wild animal.
She wasn’t afraid that he could force her to marry him,
but she was afraid of what he might say or do if he
were further defied.

“Fenn,” she said, gently, “Fenn, dear—”

“Don’t ‘Fenn, dear’ me unless you mean it! Don’t think

you can placate me by soft words that mean nothing!
Will you marry me, now?”

“I will not,” Elsie’s hauteur was the last straw.

“Then, you’ll stay here until you will!”

Whiting flung himself into a chair, and looked at her as if

he held the whip-hand.

“What do you mean?” Elsie said, icily.

“These are my rooms. You are locked in here with me, 238
alone. How long must you stay here before you decide
it’s wiser to be my wife than—”

The look the girl gave him made him quail.

“Elsie,” he said, more gently.

“Hush! Don’t dare to speak to me again. Let me out!”

She flew to the door, but it was locked, the key in

Whiting’s pocket, or the spring catch holding it, she
didn’t know which. She pounded on the door, with her
soft hands, but made little commotion that way.

“Useless, my dear,” Whiting said, calmly. “These rooms

are on a wing containing but few guests. Nobody will
hear you. Pound away, if you like.”
This wasn’t true; as a matter of fact, Whiting was very
much afraid somebody would hear her, but he deemed
this the best way to stop her,—and it was.

Elsie believed him and quit pounding. Nor did she

scream. An idea had come to her. Whiting had said
rooms. Therefore there was more to the suite than the
one they were in. Covertly she glanced at the doors,
and decided that while one rather narrow one was
doubtless a closet, the wide one, the other side of the
room, probably opened into an adjoining room, which
was likely to give on the hall.

At any rate, it was worth trying.

Cleverly, she seemed not to be noticing these details, 239

but sat, her handkerchief to her eyes, apparently
subdued and dismayed. And, in fact she was both, but
not to the point of surrender, as she appeared to
Whiting’s anxious watchfulness.

Cautiously looking about, with seemingly a vacant stare,

she saw many little personal belongings, that convinced
her the room was Whiting’s sitting room. Doubtless the
next was his bedroom. All the same, she determined to
dash through it in an attempt at freedom. If she were
quick, and the other hall door not locked, she could get
to the hall,—while if she were trapped in the other
room, her plight would be no worse than it was at
present.

She rose and walked disconsolately about,—looked from

the windows, stared, unseeing, at a picture on the wall,
—and generally appeared to be aimlessly wandering,
while she thought matters over.
Whiting watched her, but so cannily did Elsie mislead his
thoughts, that he didn’t notice she drew nearer and
nearer the bedroom door.

At last, she was almost against it, her eyes fastened on

a small clock which stood on a table at the opposite side
of the room.

“What time is it?” she said, dully, as if her decision

depended on the flight of the hours.

The ruse succeeded. He followed the direction of her 240

straining eyes, and looked at the little clock instead of
taking out his own watch.

Like a flash, Elsie tore open the door, found that it

opened into a bedroom, with a hall door, and crossing
the room in the fewest possible steps, wrenched open
the hall door. It was not locked, and she flew through it
and down the corridor toward the elevators, of which
there were two side by side.

Elsie pushed the bell so violently, that the car came up

immediately and she sprang into it, just as Whiting
came racing down the hall after her.

He rang, a long steady ring, and though Elsie’s prayers

persuaded the girl in the car with her not to go up
again, the other car shot past them flying upward.

And now Elsie achieved a master-stroke. Thinking

swiftly, she knew Whiting would make the other car
drop without a stop, and would await her on the ground
floor.

Determined to outwit him, she ordered the girl to stop

between floors and change gowns with her.
Willing enough, when Elsie offered her all the money in
her bag, and also told her she would be aiding a crime if
she refused, the little elevator girl slipped out of her
uniform, Elsie dropped off her own gown and in two
minutes they were transformed, even the cap of the girl
in place of Elsie’s pretty hat, and the hat on the other’s
head.

A little bewildered the girl then ran her car on down, 241
without stop.

At the ground floor, acting at Elsie’s orders, the other

girl stepped from the car in a furtive, hunted manner,
and ran swiftly down a long cross hall,—Whiting, full tilt
after her.

Elsie, meanwhile, stepped briskly out the front door,

sprang into a taxicab and was whirled away.

Elsie’s spirits rose. She had outwitted Fenn Whiting, and

she had escaped from a situation more dangerous than
that of the deserted taxicab of a few days before.

She went straight back to the hotel where she and the
nurse had been staying. Here the desk clerk told her
that the nurse had packed up everything and had
returned to New York.

Elsie was amazed. She trusted the nurse absolutely, but

she now began to fear her sincerity. To the poor girl it
seemed as if there were nobody in whom she could
place confidence. And there was the ever dreadful
question of the fortune. Had it not been for her insistent
family, she would have given up all thought of the
money and would have run away to hide by herself until
her birthday had passed.
But, she argued, this was not the way to feel. For she
must be at home, in case Kimball should somehow
miraculously appear.

Unable to fathom the meaning of the nurse’s departure, 242

though since she had taken all their luggage, Elsie
couldn’t think she was honest, she concluded to go right
back to New York herself.

She couldn’t hope to escape Fenn Whiting’s presence

much longer, for having learned the trick played on him,
he would of course come at once to The Turrets.

Moreover, Elsie was attracting curious looks, and even

disapproving ones by reason of her standing about in
the hall, dressed in the uniform of an elevator girl! She
wondered what the poor girl was doing, who now wore
her clothes. Perhaps she would lose her position! Elsie
determined to look after her as soon as she could
secure and count on her own safety.

And now a new dilemma presented itself. She had no

money!

All she had carried with her, in her handbag, she had
given to the girl in the elevator, thinking she would go
back to the hotel where she had her check book.

But that was gone with her trunks. Even the unpaid
cabman was already clamouring for his fare!

“Why did Miss Loring say she left?” she asked the clerk.

“She said you had sent her word you had already gone 243
home, and she was to follow at once,” he returned,
glancing at her severely. “She packed quickly and
caught the first train she could get.”
“She paid the bill?”

“Yes, in full to the time of her leaving.”

“I will ask you then, to pay this cabman, and let me

have money enough to get to New York. I will send you
a check from there.”

But the desk clerk didn’t seem to care for this plan at
all. He paid the cabman, who was becoming a nuisance,
but he declined to advance money to such an erratic
person as the lady before him seemed to be.

She had made no explanation of her strange garb, and

his manner had so roused her indignation that she kept
her own counsel.

But she was at her wits’ end. It was after four in the
afternoon and a hotel who wouldn’t lend a few dollars,
would doubtless object to her re-registering there, with
no money, and in most eccentric costume.

As she thought it over a man approached and asked if

he might be of assistance.

It was the man of the gold-filled teeth!

Any fear of him she might have felt vanished in a

strange sense of seeing an old friend! For so helpless
and friendless was the poor child that even this man,
presumably one of the “villains,” seemed a godsend!

And he was polite and deferential.

“Well,” she said, her poise returning, “all things 244

considered, I think I am privileged to ask you for the
loan of a few dollars.”
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookmass.com