Download Full Version ebook - Visit ebooknice.

com

(Ebook) Bayesian Prediction and Adaptive Sampling

Algorithms for Mobile Sensor Networks: Online
Environmental Field Reconstruction in Space and
Time by Yunfei Xu, Jongeun Choi, Sarat Dass,
Tapabrata Maiti (auth.) ISBN 9783319219202,
https://ebooknice.com/product/bayesian-prediction-and-
3319219200
adaptive-sampling-algorithms-for-mobile-sensor-networks-
online-environmental-field-reconstruction-in-space-and-
time-5235766

Click the button below to download

DOWLOAD EBOOK

Discover More Ebook - Explore Now at ebooknice.com

 Instant digital products (PDF, ePub, MOBI) ready for you
Download now and discover formats that fit your needs...

Start reading on any device today!

(Ebook) Biota Grow 2C gather 2C cook by Loucas, Jason;

Viles, James ISBN 9781459699816, 9781743365571,
9781925268492, 1459699815, 1743365578, 1925268497
https://ebooknice.com/product/biota-grow-2c-gather-2c-cook-6661374

ebooknice.com

(Ebook) SAT II Success MATH 1C and 2C 2002 (Peterson's SAT

II Success) by Peterson's ISBN 9780768906677, 0768906679

https://ebooknice.com/product/sat-ii-success-
math-1c-and-2c-2002-peterson-s-sat-ii-success-1722018

ebooknice.com

(Ebook) Matematik 5000+ Kurs 2c Lärobok by Lena

Alfredsson, Hans Heikne, Sanna Bodemyr ISBN 9789127456600,
9127456609
https://ebooknice.com/product/matematik-5000-kurs-2c-larobok-23848312

ebooknice.com

(Ebook) Master SAT II Math 1c and 2c 4th ed (Arco Master

the SAT Subject Test: Math Levels 1 & 2) by Arco ISBN
9780768923049, 0768923042
https://ebooknice.com/product/master-sat-ii-math-1c-and-2c-4th-ed-
arco-master-the-sat-subject-test-math-levels-1-2-2326094

ebooknice.com
(Ebook) Cambridge IGCSE and O Level History Workbook 2C -
Depth Study: the United States, 1919-41 2nd Edition by
Benjamin Harrison ISBN 9781398375147, 9781398375048,
1398375144, 1398375047
https://ebooknice.com/product/cambridge-igcse-and-o-level-history-
workbook-2c-depth-study-the-united-states-1919-41-2nd-edition-53538044

ebooknice.com

(Ebook) Handbook of Sensor Networks: Algorithms and

Architectures by Ivan Stojmenovic ISBN 9780471684725,
0471684724
https://ebooknice.com/product/handbook-of-sensor-networks-algorithms-
and-architectures-983016

ebooknice.com

(Ebook) Algorithms for Sensor Systems: 15th International

Symposium on Algorithms and Experiments for Wireless
Sensor Networks, ALGOSENSORS 2019, Munich, Germany,
September 12–13, 2019, Revised Selected Papers by Falko
https://ebooknice.com/product/algorithms-for-sensor-systems-15th-
Dressler, Christian Scheideler ISBN 9783030344047,
international-symposium-on-algorithms-and-experiments-for-wireless-
sensor-networks-algosensors-2019-munich-germany-
9783030344054, 3030344045, 3030344053
september-1213-2019-revised-selected-papers-10801436
ebooknice.com

(Ebook) Ad-hoc, Mobile, and Wireless Networks: 14th

International Conference, ADHOC-NOW 2015, Athens, Greece,
June 29 -- July 1, 2015, Proceedings by Symeon
Papavassiliou, Stefan Ruehrup (eds.) ISBN 9783319196619,
https://ebooknice.com/product/ad-hoc-mobile-and-wireless-
3319196618
networks-14th-international-conference-adhoc-now-2015-athens-greece-
june-29-july-1-2015-proceedings-5141724
ebooknice.com

(Ebook) Space-Time Adaptive Processing for Radar by J. R.

Guerci ISBN 9781608078202, 1608078205

https://ebooknice.com/product/space-time-adaptive-processing-for-
radar-6810434

ebooknice.com
SPRINGER BRIEFS IN ELEC TRIC AL AND COMPUTER
ENGINEERING  CONTROL, AUTOMATION AND ROBOTICS

Yunfei Xu
Jongeun Choi
Sarat Dass
Tapabrata Maiti

Bayesian Prediction
and Adaptive Sampling
Algorithms for Mobile
Sensor Networks
Online Environmental
Field Reconstruction in
Space and Time

123
SpringerBriefs in Electrical and Computer
Engineering

Control, Automation and Robotics

Series editors
Tamer Başar
Antonio Bicchi
Miroslav Krstic
More information about this series at http://www.springer.com/series/10198
Yunfei Xu Jongeun Choi Sarat Dass
• •

Tapabrata Maiti

Bayesian Prediction
and Adaptive Sampling
Algorithms for Mobile
Sensor Networks
Online Environmental Field Reconstruction
in Space and Time

123
Yunfei Xu Sarat Dass
Michigan State University Department of Statistics
East Lansing, MI Michigan State University
USA East Lansing, MI
USA
Jongeun Choi
Michigan State University Tapabrata Maiti
East Lansing, MI Department of Statistics
USA Michigan State University
East Lansing, MI
USA

ISSN 2191-8112 ISSN 2191-8120 (electronic)

SpringerBriefs in Electrical and Computer Engineering
ISSN 2192-6786 ISSN 2192-6794 (electronic)
SpringerBriefs in Control, Automation and Robotics
ISBN 978-3-319-21920-2 ISBN 978-3-319-21921-9 (eBook)
DOI 10.1007/978-3-319-21921-9

Library of Congress Control Number: 2015950872

Springer Cham Heidelberg New York Dordrecht London

© The Author(s) 2016
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made.

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
To our loving parents and beautiful families
Preface

We have witnessed a surge of applications using static or mobile sensor networks

interacting with uncertain environments. To treat a variety of useful tasks such as
environmental monitoring, adaptive sampling, surveillance, and exploration, this
book introduces a class of problems and efficient spatio-temporal models when
scalar fields need to be predicted from noisy observations collected by mobile
sensor networks. The book discusses how to make inference from the observations
based on the proposed models and also explores adaptive sampling algorithms for
robotic sensors to maximize the prediction quality subject to constraints on mem-
ory, communication, and mobility.
The objective of the book is to provide step-by-step progress in chapters for
readers to gain better understanding of the interplay between all the essential
constituents such as resource-limited mobile sensor networks, spatio-temporal
models, data-driven prediction, prediction uncertainty, and adaptive sampling for
making better predictions. The book builds on previous collective works by the
authors and is not meant to provide a comprehensive review of the topics of
interest. Specifically, materials from the previous publications by the authors [1–5]
make up a large portion of the book.
In this book, a spatio-temporal scalar field is used to represent the collection of
scalar quantities of interest, such as chemical concentration or biomass of algal
blooms (e.g., see Fig. 1.3), transported via physical processes. To deal with com-
plexity and practicality, phenomenological and statistical modeling techniques are
used to make inference from noisy observations collected, taking into account a
large scope of uncertainties. To this end, nonparametric models such as Gaussian
processes and Gaussian Markov random fields (GMRFs), along with their predic-
tion and adaptive sampling algorithms, will be explored and tailored to our needs.
The importance of selecting a Gaussian process prior via hyperparameters for given
experimental observations is illustrated (Chap. 3). Adaptive sampling to improve
the quality of hyperparameters is proposed (Chap. 3). Memory efficient prediction
based on truncated observations in space and time as well as the collective mobility
based on distributed navigation are discussed (Chap. 4). While the book starts with

vii
viii Preface

a rather simple empirical Bayes approach (Chap. 3), as we move through further
chapters, we discuss recent efforts with a fully Bayesian perspective to maximize
the flexibility of the models under various uncertainties while minimizing the
computational complexity (Chaps. 5 and 7). A fully Bayesian framework is adopted
here as it offers several advantages when inferring parameters and processes from
highly complex models (Chaps. 5 and 7). The Bayesian approach requires prior
distributions to be elicited for model parameters that are of interest. Once the priors
are elicited, the Bayesian framework is flexible and effective in incorporating all
uncertainties as well as information (limited or otherwise from data) into a single
entity, namely, the posterior. The fully Bayesian approach thus allows additional
sources and extent of uncertainties to be integrated into the inferential framework,
with the posterior distribution effectively capturing all aspects of uncertainties
involved. Subsequently, the practitioner needs only to focus on different compo-
nents of the posterior to obtain inference separately for the parameters of interest,
nuisance parameters, and hyperparameters. The fully Bayesian approach also
allows data to select the most appropriate values for nuisance parameters and
hyperparameters automatically and achieve optimal inference and prediction for the
scalar field. In this book, a fully Bayesian approach for spatio-temporal Gaussian
process regression will be formulated for resource-constrained robotic sensors to
fuse multifactorial effects of observations, measurement noise, and prior distribu-
tions for obtaining the predictive distribution of a scalar environmental field of
interest. Traditional Markov Chain Monte Carlo (MCMC) methods cannot be
implemented on resource-constrained mobile sensor networks due to high com-
putational complexity. To deal with complexity, the Bayesian spatio-temporal
models will be carefully tailored (Chap. 5). For example, we will approximate a
Gaussian process with a GMRF for computational efficiency (Chaps. 6 and 7).
A new spatial model is proposed via a GMRF (Chap. 6). In addition, ways to
improve computational efficiency will be proposed in form of empirical Bayes and
approximate Bayes instead of MCMC-based computation. For some special cases,
the developed centralized algorithms will be further refined in a distributed manner
such that mobile robots can implement distributed algorithms only using local
information available from neighboring robots over a proximity communication
graph (Chaps. 4–6).
We note that although regression problems for sensor networks under location
uncertainty have practical importance, they are not considered in this book. The
interested reader is referred to [6, 7] (centralized scheme) and [8] (distributed
scheme) for further information on this topic.

Organization

This book is organized as follows: Chapter 1 gives some background information

and a summary for each chapter. In Chap. 2, we introduce the basic mathematical
notation that will be used throughout the book. We then describe the general
Preface ix

Chapter 1: Background: Mobile sen-

sor network (MSN) and robotic sensors

Chapter 2: Preliminaries: Gaussian process (GP)

and Gaussian Markov random field (GMRF)

Chapter 3: GP, prediction, adaptive

sampling, & empirical Bayes appr.

Chapter 4: GP, prediction, & distributed strategy

Chapter 5: GP, prediction, adaptive sam-

pling, MCMC appr., & fully Bayesian appr.

Chapter 6: new spatial model, GMRF, predic-

tion, distributed strategy, & empirical Bayes appr.

Chapter 7: approximated GP, GMRF, predic-

tion, adaptive sampling, & fully Bayesian appr.

Fig. 1 Organization of chapters along with keywords

Gaussian process and its usage in nonparametric regression problems. The notations
for mobile sensor networks are also introduced in Chap. 2. In Chap. 3, we deal with
the case where hyperparameters in the covariance function is deterministic but
unknown. We design an optimal sampling strategy to improve the maximum
likelihood estimation of these hyperparameters. In Chap. 4, we assume the
hyperparameters in the covariance function are given; they can be obtained using
the approach proposed in Chap. 3. We then analyze the error bounds of prediction
error using Gaussian process regression with truncated observations. Inspired by the
error analysis, we propose both centralized and distributed navigation strategies for
mobile sensor networks to move in order to reduce prediction error variances at
points of interest. In Chap. 5, we consider a fully Bayesian approach for Gaussian
process regression in which the hyperparameters are treated as random variables.
Using discrete prior probabilities and compactly supported kernels, we provide a
way to design sequential Bayesian prediction algorithms that can be computed in
constant time as the number of observations increases. To cope with the compu-
tational complexity brought by using standard Gaussian processes with covariance
functions, in Chap. 6, we exploit the sparsity of the precision matrix by using
Gaussian Markov random fields (GMRFs). We first introduce a new class of
Gaussian processes with built-in GMRF and show its capability of representing a
wide range of nonstationary physical processes. We then derive the formulas for
x Preface

predictive statistics and design sequential prediction algorithms with fixed com-
plexity. In Chap. 7, we consider a discretized spatial field that is modeled by a
GMRF with unknown hyperparameters. From a Bayesian perspective, we design a
sequential prediction algorithm to exactly compute the predictive inference of the
random field. An adaptive sampling strategy is also designed for mobile sensing
agents to find the most informative locations in taking future measurements in order
to minimize the prediction error and the uncertainty in the estimated hyperpara-
meters simultaneously.
Keywords for chapters are summarized in Fig. 1. While each chapter is
self-contained and so can be read independently, arrows in Fig. 1 recommend
possible reading sequences for readers.

Acknowledgments

We would like to thank Songhwai Oh at Seoul National University for his

suggestions and contribution to Chap. 4. We also thank the National Science
Foundation RET teacher, Alexander Robinson, undergraduate student David York,
and Ph.D. student Huan N. Do at Michigan State University for collecting the
experimental data using a robotic boat in Chap. 3. We thank Jeffrey W. Laut,
Maurizio Porfiri (NYU Polytechnic School of Engineering), Xiaobo Tan (Michigan
State University), and Derek A. Paley (University of Maryland) for providing
pictures of their robots used in the introduction of Chap. 1.
The authors Yunfei Xu and Jongeun Choi have been supported in part by the
National Science Foundation through CAREER Award CMMI-0846547. This
support is gratefully acknowledged. Any opinions, findings, and conclusions, or
recommendations expressed in this book are those of the authors and do not
necessarily reflect the views of the National Science Foundation.

Santa Clara, California Yunfei Xu

East Lansing, Michigan Jongeun Choi
Seri Iskandar, Malaysia Sarat Dass
East Lansing, Michigan Tapabrata Maiti
May 2015
Visit https://ebooknice.com to
discover a wide range of
eBooks across various genres.
Enjoy exclusive deals and
discounts to enhance your
reading experience. Start your
digital reading journey today!
Contents

1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Contents in Chapters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1 Mathematical Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Physical Process Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Gaussian Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 Spatiotemporal Gaussian Process . . . . . . . . . . . . . . . . . . 13
2.2.3 Gaussian Markov Random Field . . . . . . . . . . . . . . . . . . 14
2.3 Mobile Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Gaussian Processes for Regression . . . . . . . . . . . . . . . . . . . . . . 17
3 Learning Covariance Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1 Selection of Gaussian Process Prior . . . . . . . . . . . . . . . . . . . . . 19
3.2 Learning the Hyperparameters . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Optimal Sampling Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 Simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4 Memory Efficient Prediction With Truncated Observations . . . . . . 27
4.1 GPR with Truncated Observations . . . . . . . . . . . . . . . . . . . . . . 28
4.1.1 Error Bounds Using Truncated Observations . . . . . . . . . . 28
4.1.2 Selecting Temporal Truncation Size . . . . . . . . . . . . . . . . 34
4.2 Optimal Sampling Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2.1 Centralized Navigation Strategy . . . . . . . . . . . . . . . . . . . 37
4.2.2 Distributed Navigation Strategy . . . . . . . . . . . . . . . . . . . 39
4.3 Simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3.1 Centralized Sampling Scheme . . . . . . . . . . . . . . . . . . . . 46
4.3.2 Distributed Sampling Scheme . . . . . . . . . . . . . . . . . . . . 48

xi
xii Contents

5 Fully Bayesian Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.1 Fully Bayesian Prediction Approach . . . . . . . . . . . . . . . . . . . . . 54
5.1.1 Prior Selection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.1.2 MCMC-Based Approach. . . . . . . . . . . . . . . . . . . . . . . . 56
5.1.3 Importance Sampling Approach . . . . . . . . . . . . . . . . . . . 61
5.1.4 Discrete Prior Distribution. . . . . . . . . . . . . . . . . . . . . . . 64
5.2 Sequential Bayesian Prediction . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2.1 Scalable Bayesian Prediction Algorithm . . . . . . . . . . . . . 65
5.2.2 Distributed Implementation for a Special Case. . . . . . . . . 66
5.2.3 Adaptive Sampling. . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3 Simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.1 MCMC-Based Approach on a 1-D Scenario . . . . . . . . . . 70
5.3.2 Centralized Scheme on 1-D Scenario . . . . . . . . . . . . . . . 71
5.3.3 Distributed Scheme on 2-D Scenario . . . . . . . . . . . . . . . 74
6 New Efficient Spatial Model with Built-In Gaussian Markov
Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.1 Spatial Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.1.1 Spatial Model Based on GMRF . . . . . . . . . . . . . . . . . . . 78
6.1.2 Gaussian Process Regression . . . . . . . . . . . . . . . . . . . . . 79
6.1.3 Sequential Prediction Algorithm. . . . . . . . . . . . . . . . . . . 82
6.2 Distributed Spatial Prediction. . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.2.1 Distributed Computation . . . . . . . . . . . . . . . . . . . . . . . . 84
6.2.2 Distributed Prediction Algorithm . . . . . . . . . . . . . . . . . . 84
6.3 Simulation and Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.3.1 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.3.2 Centralized Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.3.3 Distributed Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.3.4 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7 Fully Bayesian Spatial Prediction Using Gaussian Markov
Random Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.1 Spatial Field Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.2 Bayesian Predictive Inference . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.3 Sequential Bayesian Inference . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.3.1 Update Full Conditional Distribution . . . . . . . . . . . . . . . 97
7.3.2 Update Likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.3.3 Update Predictive Distribution . . . . . . . . . . . . . . . . . . . . 99
7.4 Adaptive Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.5 Simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

Appendix A: Mathematical Background. . . . . . . . . . . . . . . . . . . . . . . 107

References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Chapter 1
Introduction

1.1 Background

Sensor networks are ubiquitous due to the recent technological breakthroughs in

micro-electro-mechanical systems (MEMS), wireless communications, and embed-
ded systems [9, 10]. A sensor network consists of a collection of low-cost, low-power,
and multifunctional sensing devices that communicate over finite distances. A flexi-
ble and application-specific operating system, TinyOS, was developed at UC Berke-
ley for sensor networks with severe memory and power constraints [11]. TinyOS
runs on small and cheap wireless sensor nodes (e.g., MICA2DOT from Crossbow
Technology, Inc., CA, USA) as shown in Fig. 1.1a. Such sensor nodes have been
equipped with various environmental and ambient sensors such as temperature sen-
sors, lighting sensors, chemical sensors, accelerometers, and RFID readers along
with the communication capability with neighbors via low-power wireless commu-
nication to form a wireless ad hoc sensor network with up to 100,000 nodes [10].
Endowing the nodes in a sensor network with mobility significantly increases the
sensor network’s sampling capabilities [12, 13]. The sensor networks which consist
of mobile sensing agents are more flexible than the ones with only static nodes. A
conceptual picture of a distributed mobile sensor network with a (R-disk) proximity
communication graph model is shown in Fig. 1.1b, which assumes that a robotic
sensor can communicate with its neighboring robots within distance R. Devised in
the Laboratory of Intelligent Systems at the Swiss Federal Institute of Technology,
the Swarming Micro Air Vehicle Network (SmavNet) depicted in Fig. 1.1c allows
a single operator to control an entire swarm of cheap unmanned aerial vehicles
(UAVs) for search and rescue operation [14]. 3D Robotics, Inc., Berkeley, USA
produces “personal drones” such as DIY drone kits as well as ready-to-fly quadrotors,
multirotors, and fixed wing UAVs based on open source UAV autopilot platforms. A
quadrotor with a camera from 3D Robotics is shown in Fig. 1.2.
Biologists and other scientists are interested in leveraging recent technological
advances [10, 15, 16] by deploying mobile sensor networks for environmental and

© The Author(s) 2016 1

Y. Xu et al., Bayesian Prediction and Adaptive Sampling Algorithms
for Mobile Sensor Networks, SpringerBriefs in Control,
Automation and Robotics, DOI 10.1007/978-3-319-21921-9_1
2 1 Introduction

(a) (b)

(c)

Fig. 1.1 a Wireless microsensor mote (MICA2DOT) from Crossbow Technology, Inc., CA, USA,
(www.xbow.com). b Distributed mobile sensor network with a (R-disk) proximity communication
graph for environmental monitoring (credit: Justin Mrkva). c Swarming Micro Air Vehicle Network
(SmavNet) developed in the Laboratory of Intelligent Systems at the Swiss Federal Institute of
Technology for search and rescue operation (Photo courtesy of Laboratory of Intelligent Systems
at EPFL, http://lis.epfl.ch)

wildlife monitoring. For example, one of the pressing societal concerns about water
quality is the proliferation of harmful algal blooms in ponds, lakes, rivers, and coastal
ocean worldwide. A satellite image of a 2011 significant harmful algal bloom in
western Lake Erie is shown in Fig. 1.3. The excessive growth of cyanobacteria leads
to a decaying biomass and oxygen depletion, which are detrimental to fish and other
aquatic life as well as to land animals and humans consequently (due to the produced
1.1 Background 3

Fig. 1.2 Personal drone (IRIS+) manufactured by 3D Robotics Inc., Berkeley, USA (credit: 3D
Robotics, http://3drobotics.com)

Fig. 1.3 Satellite image of 2011 significant harmful algal bloom in western Lake Erie in
Michigan, which impacted over half of the lake shore (credit: MERIS/ESA, processed by
NOAA/NOS/NCCOS, http://www.noaanews.noaa.gov)

toxins that deteriorate water quality) [17–19]. Deploying mobile sensor networks
can be a viable way to reconstruct and monitor such harmful algal blooms [20, 21].
Indeed, we have seen the increasing exploration of robotic technologies in aquatic
sensing [20, 22–25]. A robotic boat was used in concert with stationary buoys to form
an aquatic microbial system [20]; spatiotemporal aquatic field reconstruction was
implemented using inexpensive, low-power, robotic fish in [21] (see also Fig. 1.5a
for gliding robotic fish [26]); and low-cost, self-sustained mobile surface vehicles
have been designed for environmental monitoring as part of the citizen science project
Brooklyn Atlantis [23] (see Fig. 1.4). A robotic boat equipped with a depth sensor, as
4 1 Introduction

(a)

(b)

Fig. 1.4 Robots developed by NYU Polytechnic School of Engineering for image and water quality
data collection as part of the citizen science project Brooklyn Atlantis (http://www.brooklynatlantis.
poly.edu) [23] (credit: Jeffrey W Laut)

shown in Fig. 1.5b, can sample the depth of a lake for its estimation [27]. Autonomous
underwater vehicles (AUVs) are being developed as an important tool in oceanog-
raphy, marine biology, and other maritime applications [28–30]. Autonomous sea
gliders are another noteworthy example. These battery-powered, buoyancy-driven
vehicles can travel thousands of miles horizontally, for many months, without chang-
ing or recharging batteries [31–34]. With the networks of gliders as shown in Fig. 1.6,
adaptive sampling has been demonstrated in Monterey Bay, California [35–37].
1.1 Background 5

(a)

(b)

Fig. 1.5 a Gliding robotic fish “Grace” sampling harmful algae in the Wintergreen Lake, Michigan
[26] (credit: Xiaobo Tan). b Robotic boat sampling depth near Hawk Island, Lansing, Michigan
(credit: Jongeun Choi)

The robotic sensor technologies have then brought an increasing exploitation of

navigation of mobile sensor networks and robotic sensors interacting with uncertain
environments [2, 35–42]. A necessity in such scenarios is to design algorithms to
process collected observations from environments (e.g., distributed estimators) for
robots such that either the local information about the environment can be used for
local control actions or the global information can be estimated asymptotically.
The approach of designing such algorithms takes two different paths depend-
ing on whether it uses an environmental model in space and time or not. Without
6 1 Introduction

Fig. 1.6 Gliders used for adaptive sampling [35–37] (credit: Derek A. Paley)

environmental models, extremum seeking control has been proven to be very effective
for finding a source of a signal (chemical, electromagnetic, etc.) [38, 39]. Distributed
algorithms for stochastic source seeking with mobile robot networks have been devel-
oped for both cases with and without the mutual information model between their
expected measurements and the expected source location [43]. A unifying frame-
work of distributed stochastic gradient algorithms that can deal with coverage control,
spatial partitioning, and dynamic vehicle routing problems in the absence of a priori
knowledge of the event location distribution has been presented in [40].
A drawback of the spatial model free approach is that it limits its task to finding
the maximum (or minimum) point of the environmental field. To tackle a variety
of useful tasks such as the exploration, estimation, prediction, and maximum seek-
ing of a scalar field, it is essential for robots to have a spatial (and temporal) field
model [2–4, 36, 41, 42, 44–50]. Although control algorithms for mobile robots have
been developed based on computationally demanding, physics-based field models
[51], for resource-constrained mobile robots, recently, phenomenological and statis-
tical modeling techniques such as kriging, Gaussian process regression, and kernel
regression have gained much attention. Among phenomenological spatial models,
adaptive control of multiple robotic sensors based on a parametric approach needs
a persistent excitation (PE) condition for convergence of parameters [42, 50], while
control strategies based on Bayesian spatial models do not require such conditions
(e.g., by utilizing priori distributions as in Kalman filtering [41] or Gaussian process
regression [2]). Hence, control engineers have become more aware of the useful-
ness of nonparametric Bayesian approaches such as Gaussian processes (defined by
1.1 Background 7

mean and covariance functions) [52, 53] to statistically model physical phenomena
for the navigation of mobile sensor networks, e.g., [2–4, 36, 44–47]. Other more
data-driven approaches have also developed (without statistical structure used in
Gaussian processes) such as using kernel regression [48] and in reproducing kernel
Hilbert spaces [49]. However, without a statistical structure in a random field, such
an approach (as in [48, 49]) usually requires a great number of observations than the
one with a statistical structure for a decent prediction quality.
In a mobile sensor network, the resource-limited sensing agents are required to
collaborate in order to achieve a specific objective. The cooperative control becomes
essential. The most popular applications are in networks of autonomous ground
vehicles [54, 55], underwater and surface vehicles [23, 36, 56–58], or aerial vehicles
[59–61]. Emerging technologies have been reported on the coordination of mobile
sensing agents [41, 62–70].
The mobility of mobile agents can be designed in order to perform the optimal
sampling of the field of interest. Optimal sampling design is the process of choosing
where to take samples in order to maximize the information gained. Recently, in
[36], Leonard et al. developed mobile sensor networks that optimize ocean sampling
performance defined in terms of uncertainty in a model estimate of a sampled field.
A typical sensor placement technique [71] that puts sensors at the locations where
the entropy is high tends to place sensors along the borders of the area of interest
[44]. In [44], Krause et al. showed that seeking sensor placements that are most
informative about unsensed locations is NP-hard, and they presented a polynomial
time approximation algorithm by exploiting the submodularity of mutual information
[72]. In a similar approach, in [73], Singh et al. presented an efficient planning of
informative paths for multiple robots that maximize the mutual information.
To find these locations that predict the phenomenon best, one needs a model
of the spatiotemporal phenomenon. To this end, we use Gaussian processes (and
Gaussian random fields) to model fields undergoing transport phenomena. Nonpara-
metric Gaussian process regression (or Kriging in geostatistics) has been widely used
as a nonlinear regression technique to estimate and predict geostatistical data [52, 53,
74, 75]. A Gaussian process is a natural generalization of the Gaussian probability
distribution. It generalizes a Gaussian distribution with a finite number of random
variables to a Gaussian process with an infinite number of random variables in the
surveillance region [53]. Gaussian process modeling enables us to efficiently pre-
dict physical values, such as temperature, salinity, pH, or biomass of harmful algal
blooms, at any point with a predicted uncertainty level. For instance, near-optimal
static sensor placements with a mutual information criterion in Gaussian processes
were proposed in [44, 76]. A distributed Kriged Kalman filter for spatial estimation
based on mobile sensor networks was developed in [45]. Multiagent systems that are
versatile for various tasks by exploiting predictive posterior statistics of Gaussian
processes were developed in [77, 78].
Gaussian process regression, based on the standard mean and covariance func-
tions, requires an inversion of a covariance matrix whose size grows as the number
of observations increases. The significant computational complexity in Gaussian
8 1 Introduction

process regression due to the growing number of observations (and hence the size of
covariance matrix) has been tackled in different ways [2, 79–83].
Unknown hyperparameters in the covariance function can be estimated by a max-
imum likelihood (ML) estimator or a maximum a posteriori (MAP) estimator and
then be used in the prediction as the true hyperparameters [1]. However, the point
estimate (ML or MAP estimate) itself needs to be identified using a sufficient amount
of measurements and it fails to incorporate the uncertainty in the estimated hyper-
parameters into the prediction in a Bayesian perspective. The advantage of a fully
Bayesian approach is that the uncertainty in the model parameters is incorporated in
the prediction [84]. In [85], Gaudard et al. presented a Bayesian method that uses
importance sampling for analyzing spatial data sampled from a Gaussian random
field whose covariance function was unknown. However, the solution often requires
Markov Chain Monte Carlo (MCMC) methods, which greatly increases the com-
putational complexity. In [46], an iterative prediction algorithm without resorting
to MCMC methods has been developed based on analytical closed-form solutions
from results in [85], by assuming that the covariance function of the spatiotemporal
Gaussian random field is known up to a constant.
There have been growing efforts to fit a computationally efficient Gaussian
Markov random field (GMRF) on a discrete lattice to a Gaussian random field on a
continuum space [4, 86–88]. It has been demonstrated that GMRFs with small neigh-
borhoods can approximate Gaussian fields surprisingly well [86]. This approximated
GMRF and its regression are very attractive for the resource-constrained mobile sen-
sor networks due to its computational efficiency and scalability [89] as compared to
the standard Gaussian process and its regression. Fast kriging of large datasets using
a GMRF as an approximation of a Gaussian field has been proposed in [88].

1.2 Contents in Chapters

A brief summary for each subsequent chapter is as follows. Chapter 2 gives an

introduction to Gaussian processes and Gaussian Markov random fields for general
domains as well as the space-time domain.
In Chap. 3, we develop covariance function learning algorithms for the sensing
agents to perform nonparametric prediction based on a properly adapted Gaussian
process for a given spatiotemporal phenomenon. By introducing a generalized covari-
ance function, we expand the class of Gaussian processes to include the anisotropic
spatiotemporal phenomena. Maximum likelihood (ML) optimization is used to esti-
mate hyperparameters for the associated covariance function as an empirical Bayes
method. The proposed optimal navigation strategy for autonomous vehicles will max-
imize the Fisher information [90], improving the quality of the estimated covariance
function.
In Chap. 4, we first present a theoretical foundation of Gaussian process regres-
sion with truncated observations. In particular, we show that the quality of prediction
based on truncated observations does not deteriorate much as compared to that of
Visit https://ebooknice.com to
discover a wide range of
eBooks across various genres.
Enjoy exclusive deals and
discounts to enhance your
reading experience. Start your
digital reading journey today!
1.2 Contents in Chapters 9

prediction based on all cumulative data under certain conditions. The error bounds to
use truncated observations are analyzed for prediction at a single point of interest. A
way to select the temporal truncation size for spatiotemporal Gaussian processes is
also introduced. Inspired by the analysis, we then propose both centralized and dis-
tributed navigation strategies for mobile sensor networks to move in order to reduce
prediction error variances at points of interest. In particular, we demonstrate that
the distributed navigation strategy produces an emergent, swarming-like, collective
behavior to maintain communication connectivity among mobile sensing agents.
In Chap. 5, we formulate a fully Bayesian approach for spatiotemporal Gaussian
process regression under practical conditions such as measurement noise and
unknown hyperparameters (particularly, the bandwidths). Thus, multifactorial effects
of observations, measurement noise, and prior distributions of hyperparameters are
all correctly incorporated in the computed posterior predictive distribution. Using
discrete prior probabilities and compactly supported kernels, we provide a way to
design sequential Bayesian prediction algorithms that can be computed (without
using the Gibbs sampler) in constant time (i.e., O(1)) as the number of observations
increases. An adaptive sampling strategy for mobile sensors, using the maximum
a posteriori (MAP) estimation, has been proposed to minimize the prediction error
variances.
In Chap. 6, we propose a new class of Gaussian processes for resource-constrained
mobile sensor networks that build on a Gaussian Markov random field (GMRF) with
respect to a proximity graph over the surveillance region. The main advantages of
using this class of Gaussian processes over standard Gaussian processes defined by
mean and covariance functions are its numerical efficiency and scalability due to
its built-in GMRF and its capability of representing a wide range of nonstationary
physical processes. The formulas for predictive statistics are derived and a sequential
field prediction algorithm is provided for sequentially sampled observations. For a
special case using compactly supported weighting functions, we propose a distributed
algorithm to implement field prediction by correctly fusing all observations.
In Chap. 7, we consider a discretized spatial field that is modeled by a GMRF with
unknown hyperparameters. From a Bayesian perspective, we design a sequential pre-
diction algorithm to exactly compute the predictive inference of the random field.
The main advantages of the proposed algorithm are (1) the computational efficiency
due to the sparse structure of the precision matrix, and (2) the scalability as the num-
ber of measurements increases. Thus, the prediction algorithm correctly takes into
account the uncertainty in hyperparameters in a Bayesian way and also is scalable to
be usable for the mobile sensor networks with limited resources. An adaptive sam-
pling strategy is also designed for mobile sensing agents to find the most informative
locations in taking future measurements in order to minimize the prediction error
and the uncertainty in the estimated hyperparameters simultaneously.
Chapter 2
Preliminaries

2.1 Mathematical Notation

Standard notation is used throughout this book. Let R, R≥0 , R>0 , Z, Z≥0 , Z>0 denote
the sets of real numbers, nonnegative real numbers, positive real numbers, integers,
nonnegative integers, and positive integers, respectively.
Let E, Var, Corr, Cov denote the expectation, variance, correlation, and the covari-
ance operators, respectively.
Let AT ∈ R M×N be the transpose of a matrix A ∈ R N ×M . Let tr(A) and det(A)
denote the trace and the determinant of a matrix A ∈ R N ×N , respectively. Let
rowi (A) ∈ R M and col j (A) ∈ R N denote the ith row and the jth column of a matrix
A ∈ R N ×M , respectively.
The positive definiteness and the positive semi-definiteness of a square matrix A
are denoted by A  0 and A  0, respectively.
Let |x| denote the absolute value of a scalar x. Let x denote the standard
Euclidean norm (2-norm) of a vector x. The induced 2-norm of a matrix A is denoted
by A. Let x∞ denote the infinity norm of a vector x.
Let 1 denote the vector with all elements equal to one and I denote the identity
matrix with an appropriate size. Let ei be the standard basis vector of appropriate
size with 1 as its ith element and 0 on all other elements.
The symbol ⊗ denotes the Kronecker product. The symbol ◦ denotes the
Hadamard product (also known as the entry-wise product and the Schur product).
A random vector x, which is distributed by a normal distribution of mean μ and
covariance matrix C, is denoted by x ∼ N(μ, C). The corresponding probability
density function is denoted by N(x; μ, C).
The relative complement of a set A in a set B is denoted by B \ A := B ∩ Ac ,
where Ac is the complement of A. For a set A ∈ I, we define z A = {z i | i ∈ A}.
Let −A denote the set I \ A.

© The Author(s) 2016 11

Y. Xu et al., Bayesian Prediction and Adaptive Sampling Algorithms
for Mobile Sensor Networks, SpringerBriefs in Control,
Automation and Robotics, DOI 10.1007/978-3-319-21921-9_2
12 2 Preliminaries

An undirected graph G = (V, E) is a tuple consisting of a set of vertices V :=

{1, · · · , n} and a set of edges E ⊂ V × V. The neighbors of i ∈ V in G are denoted
by Ni := { j ∈ V | {i, j} ∈ E}.
Other notations will be explained in due course.

2.2 Physical Process Model

In this section, we review important notions for the Gaussian process which will be
used to model the physical phenomenon. In particular, we introduce a class of spa-
tiotemporal Gaussian process model with anisotropic covariance functions. The prop-
erties of Gaussian Markov random fields (GMRF) are also briefly reviewed.

2.2.1 Gaussian Process

A Gaussian process can be thought of a generalization of a Gaussian distribution over

a finite vector space to function space of infinite dimension. It is formally defined as
follows [53, 91]:

Definition 2.1 A Gaussian process (GP) is a collection of random variables, any

finite number of which have a consistent1 joint Gaussian distribution.

A Gaussian process, denoted by

 
z(x) ∼ GP μ(x), C(x, x ; θ ) (2.1)

is completely specified by its mean function μ(x) and covariance function C(x, x ; θ )
which are defined as

μ(x) = E [z(x)] ,
 
C(x, x ; θ ) = E (z(x) − μ(x)) (z(x ) − μ(x ))|θ .

Although not needed to be done, we take the mean function to be zero for notational
simplicity,2 i.e., μ(x) = 0. If the covariance function C(x, x ; θ ) is invariant with
respect to translations in the input space, i.e., C(x, x ; θ ) = C(x − x ; θ ), we call it
stationary. Furthermore, if the covariance functionis a function of only the distance
between the inputs, i.e., C(x, x ; θ ) = C(x − x  ; θ ), then it is called isotropic.

1 Itis also known as the marginalization property. It means simply that the random variables obey
the usual rules of marginalization, etc.
2 This is not a drastic limitation since the mean of the posterior process is not confined to zero [53].
2.2 Physical Process Model 13

Fig. 2.1 Realization of a

two-dimensional (D = 2)
Gaussian process with
σ 2f = 5, σ1 = 2.5, and
σ2 = 1.5.

In practice, a parametric family of functions is used instead of fixing the covariance

function [84]. One common choice of a stationary covariance function is
 D  2
 x − x
C(x, x ; θ ) = σ 2f exp − , (2.2)
=1
2σ2

where x is the th element of x ∈ R D . From (2.2), it can be easily seen that the
correlation between two inputs decreases as the distance between them increases.
This decreasing rate depends on the choice of the length scales {σ }. A very large
length scale means that the predictions would have little bearing on the correspond-
ing input which is then said to be insignificant. σ 2f gives the overall vertical scale
relative to the mean of the Gaussian process in the output space. These parame-
ters play the role of hyperparameters since they correspond to the hyperparame-
ters in neural networks and in the standard parametric model. Therefore, we define
θ = (σ 2f , σ1 , · · · , σ D )T ∈ R D+1 as the hyperparameter vector. A realization of a
Gaussian process that is numerically generated is shown in Fig. 2.1.

2.2.2 Spatiotemporal Gaussian Process

In this section, spatiotemporal Gaussian processes are of particular interest. Spa-

tiotemporal Gaussian processes are obtained as a special case of (2.1) by setting
x ⊂ R D × R≥0 , where R D is for spatial locations and R≥0 is the temporal domain.
A spatiotemporal Gaussian process can be written as

z(s, t) ∼ GP(μ(s, t), C(s, t, s , t ; θ )),

14 2 Preliminaries

(a) (b) (c)

Fig. 2.2 Realization of a spatiotemporal (D = 2) Gaussian process with σ 2f = 5, σ1 = 2.5,

σ2 = 1.5, and σt = 8 at a t = 1, b t = 5, and c t = 10.

where x = (sT , t)T ∈ R D × R≥0 . We consider the following generalized

anisotropic covariance function C(x, x ; θ ) with a hyperparameter vector θ :=
(σ 2f , σ1 , · · · , σ D , σt )T ∈ R D+2 :


D
(s − s )2 (t − t )2
C(x, x ; θ ) = σ 2f exp − exp − , (2.3)
=1
2σ2 2σt2

where s, s ∈ Q ⊂ R D , t, t ∈ R≥0 . {σ1 , · · · , σ D } and σt are kernel bandwidths

for space and time, respectively. (2.3) shows that points close in the measurement
space and time indices are strongly correlated and produce similar values. In reality,
the larger temporal distance two measurements are taken with, the less correlated
they become, which strongly supports our generalized covariance function in (2.3).
This may also justify the truncation (or windowing) of the observed time series data
to limit the size of the covariance matrix for reducing the computational cost. A
spatially isotropic version of the covariance function in (2.3) has been used in [36].
A realization of a spatiotemporal Gaussian process that is numerically generated is
shown in Fig. 2.2.

2.2.3 Gaussian Markov Random Field

The Gaussian Markov random field is formally defined as follows [92]:

Definition 2.2 (GMRF, [92, Definition 2.1]) A random vector z = (z 1 , · · · , z N )T ∈

R N is called a GMRF with respect to a graph G = (V, E) with mean μ and precision
matrix Q  0, if and only if its density has the form

|Q|1/2 1
π(z) = /2
exp − (z − μ)T Q(z − μ) ,
(2π ) N 2
2.2 Physical Process Model 15

and (Q)i j = 0 ⇔ {i, j} ∈ E for all i = j, where the precision matrix (or information
matrix) Q = C−1 is the inverse of the covariance matrix C, and |Q| denotes the
determinant of Q.

The Markov property of a GMRF can be shown by the following theorem.

Theorem 2.1 ([92, Theorem 2.4]) Let z be a GMRF with respect to G = (V, E).
Then the followings are equivalent.
1. The pairwise Markov property:

z i ⊥z j | z −i j if {i, j} ∈
/ E and i = j,

where ⊥ denotes conditional independence and z −i j := z −{i, j} = z I \{i, j} . This

implies that z i and z j are conditionally independent given observations at all
other vertices except {i, j} if i and j are not neighbors.
2. The local Markov property:

z i ⊥z −{i,Ni } | z Ni for every i ∈ I.

3. The global Markov property:

z A ⊥z B | z C

for disjoint sets A, B, and C where C separates A and B, and A and B are
nonempty.
If a graph G has small cardinalities of the neighbor sets, its precision matrix Q
becomes sparse with many zeros in its entries. This plays a key role in computation
efficiency of a GMRF which can be greatly exploited by the resource-constrained
mobile sensor network. For instance, some of the statistical inference can be obtained
directly from the precision matrix Q with conditional interpretations.
Theorem 2.2 ([92, Theorem 2.3]) Let z be a GMRF with respect to G = (V, E)
with mean μ and precision matrix Q  0, then we have

1 
E(z i | z −i ) = μi − (Q)i j (z j − μ j ),
(Q)ii
j∈Ni
1
Var(z i | z −i ) = ,
(Q)ii
(Q)i j
Corr(z i , z j | z −i j ) = −  , ∀i = j.
(Q)ii (Q) j j
16 2 Preliminaries

2.3 Mobile Sensor Network

In this section, we explain the sensor network formed by multiple mobile sensing
agents and present the measurement model used throughout the thesis.
Let N be the number of sensing agents distributed over the surveillance region
Q ∈ R D . The identity of each agent is indexed by I := {1, 2, · · · , N }. Assume that
all agents are equipped with identical sensors and take noisy observations at time
t ∈ Z>0 . At time t, the sensing agent i takes a noise-corrupted measurement yi (t)
at its current location qi (t) ∈ Q, i.e.,

i.i.d.
yi (t) = z(qi (t), t) + i , i ∼ N(0, σw2 ),

where the sensor noise i is considered to be an independent and identically dis-

tributed Gaussian random variable. σw2 > 0 is the noise level and we define the
signal-to-noise ratio as
σ 2f
γ = 2.
σw

Notice that when a static field is considered, we have z(s, t) = z(s).

For notational simplicity, we denote the collection of positions of all N agents at
time t as q(t), i.e.,
 T
q(t) := q1 (t)T , · · · , q N (t)T ∈ QN .

The collective measurements from all N mobile sensors at time t are denoted by

yt := (y1 (t), · · · , y N (t))T ∈ R N .

The cumulative measurements from time t ∈ Z>0 to time t ∈ Z>0 are denoted by
 T
yt:t := ytT , · · · , ytT ∈ R N (t −t+1) .

The communication network of mobile agents can be represented by an undirected

graph. Let G(t) := (I, E(t)) be an undirected communication graph such that an edge
(i, j) ∈ E(t) if and only if agent i can communicate with agent j = i at time t. We
define the neighborhood of agent i at time t by Ni (t) := { j ∈ I | (i, j) ∈ E(t)}.
Similarly, let q[i] (t) denote the vector form of the collection of positions in
 
q j (t) | j ∈ {i} ∪ Ni (t) . Let yt[i] denote vector form of the collection of obser-
 
vations in y(q j (t), t) | j ∈ {i} ∪ Ni (t) . The cumulative measurements of agent i
[i]
from time t to time t are denoted as yt:t .
2.4 Gaussian Processes for Regression 17

2.4 Gaussian Processes for Regression

 
Suppose we have a dataset D = (x(i) , y (i) ) | i = 1, · · · , n collected by mobile
sensing agents where x(i) denotes an input vector of dimension D and y (i) denotes a
scalar value of the noise-corrupted output. The objective of probabilistic regression
is to compute the predictive distribution of the function values z ∗ := z(x∗ ) at some
test input x∗ .
For notational simplicity, we define the design matrix X of dimension n × D as
the aggregation of n input vectors (i.e., rowi (X) := (x(i) )T ), and the outputs are
collected in a vector y := (y (1) , · · · , y (n) )T . The corresponding vector of noise-free
outputs is defined as z := (z(x(1) ), · · · , z(x(n) ))T .
The advantage of the Gaussian process formulation is that the combination of the
prior and noise models can be carried out exactly via matrix operations [93]. The
idea of Gaussian process regression is to place a GP prior directly on the space of
functions without parameterizing the function z(·), i.e.,

π(z|θ) = N(z; μ, K),

where μ ∈ Rn is the mean vector obtained by (μ)i = μ(x(i) ), and K :=

Cov(z, z|θ) ∈ Rn×n is the covariance matrix obtained by (K)i j = C(x(i) , x( j) ; θ ).
Notice that the GP model and all expressions are always conditional on the corre-
sponding inputs. In the following, we will always neglect the explicit conditioning
on the input matrix X.
The inference in the Gaussian process model is as follows. First, we assume a
joint GP prior π(z, z ∗ |θ ) over functions, i.e.,
   
μ K k
π(z, z ∗ |θ) = N , T , (2.4)
μ(x∗ ) k C(x∗ , x∗ ; θ )

where k := Cov(z, z ∗ |θ ) ∈ Rn is the covariance between z and z ∗ obtained by

(k)i = C(x(i) , x∗ ; θ ). Then, the joint posterior is obtained using Bayes rule, i.e.,

π(y|z)π(z, z ∗ |θ )
π(z, z ∗ |θ, y) = ,
π(y|θ)

where we have used π(y|z, z ∗ ) = π(y|z). Finally, the desired predictive distribution
π(z ∗ |θ, y) is obtained by marginalizing out the latent variables in z, i.e.,

π(z ∗ |θ , y) = π(z, z ∗ |θ, y)dz
 (2.5)
1
= π(y|z)π(z, z ∗ |θ, y)dz.
π(y|θ)
18 2 Preliminaries

Since we have the joint Gaussian prior given in (2.4) and

 
y|z ∼ N z, σw2 I ,

the integral in (2.5) can be evaluated in closed-form and the predictive distribution
turns out to be Gaussian, i.e.,
 
z ∗ |θ , y ∼ N μz ∗ |θ,y , σz2∗ |θ,y , (2.6)

where
μz ∗ |θ,y = μ(x∗ ) + k T (K + σw2 I)−1 (y − μ), (2.7)

and
σz2∗ |θ,y = C(x∗ , x∗ ; θ ) − k T (K + σw2 I)−1 k. (2.8)

For notational simplicity, we define the covariance matrix of the noisy observations
as C := Cov(y, y|θ) = K + σw2 I.
Random documents with unrelated
content Scribd suggests to you:
Salisbury, W., 112
Saltwood, R., 117, 119;
Comparison between four birds, J. Mychell, 118
Sandars, Samuel, 110, 128
Sandford, J., 99
Sawtry, J., Defence of the marriage of priests, Jan Troost, 115
Scauseby, Sir J., 50
Scolar, J., 67-70, 75, 100
Scoloker, J., 106, 107, 110, 111
Selden, J., 78
Seres, W., 107
Seville, Bibliotheca Columbiniana, 45
Short epistle to such as condemn the marriage of priests, J.
Mychell, 118
Shrewsbury stationer, 112
Siberch, J., 73-80, 84-89, 95, 96
Signs:
Arma Regia, Royal Arms, J. Siberch, 76, 79
Bull, F. Freez, 44
Elephant, F. Regnault, 61
Long Shop in the Poultry, J. Mychell, 117
Our Lady of Pity, W. de Worde, 58
Rose, F. Freez, 44
St John Evangelist, G. Chastelayn, 25
Sun, W. de Worde, 57, 58
Trinity, H. Jacobi, 26
Singer, S. W., 7
Sirectus, A., Formalitates, W. de Worde, 26, 68
Smith, George, 5
Smith, Richard, sale, 65
Solen, A., 125
Southwark, printing, 28
Sparchford, R., 81
Spencer, G. J. Earl, 5, 10
Speryng, N., 80, 84, 86-88, 93-97
Spire, John of. See John
Spiritual matrimony between Christ and the church, J. Oswen, 113
Spiritual matrimony between Christ and the soul, J. Mychell, 118
Squire, W., 89, 90
Stanbridge, J., 13;
Accidence, H. Goes, 53;
J. Mychell, 118;
R. Pynson, 25;
Vocabula, M. Coffin, 120
Stationer to the King, 23
Stationers, University, Cambridge, 89, 90;
Oxford, 22
Statutes of 7 Edward VI., J. Oswen, 114, 115
Statutes of the Stannary, 99
Stevenage, R., 101-103
Straten, D. van der, 108, 109
Strype, J., 71, 72
Sturbridge fair, 126
Sutherland, Duke of, sale, 28
Swyneshede, R., Logica, Oxford, 14, 16
Symonds, T., 92
Symson, Sir John, 50

Tabbe, H., 103

Tanner, T., 71
Ten Raem, Gerard, 8
Terentius, P., 13;
Vulgaria Terentii, Oxford, 14, 15
Testament, New, Latin, 86, 87;
English, J. Oswen, 112, 113
Theodoric, 19, 20
Ther Hoernen, A., 16, 20
Thoresby, Ralph, 43
Thorne, J. See Dorne
Thorpe, T., 105
Tonstall, C., Oratio in laudem matrimonii Mariæ et Francisci, R.
Pynson, 68
Tractatus secundarum intentionum logicalium, P. Treveris, 29
Treatise of baptism and the Lord’s Supper, Grunning, 115
Treveris, P., 28, 29
Trinity College. See Cambridge, Dublin
Troost, Jan, 115
Turner, Robert, 3
Turner, W., Rescuynge of the Romish fox, H. Hitprick, 120
Tutet, M. C., 36, 57

Uffington, T., 23
Upton, N., 40
Van Damme, P., 5
Van der Haeghen, G., 87
Van der Straten. See Straten
Van Ruremond. See Ruremond
Vaughan, R., 45, 66
Vellum, books printed on, 11, 12, 17, 38, 44, 59, 81, 83
Venerie de Twety, 40
Venice, printing, 18, 43, 77
Veron, J., 113, 114;
Godly sayings of the old fathers, J. Oswen, 113
Very declaration of the bond and free will of man, J Herford, 102
Violette, Pierre, 45, 46
Virgilius Maro, P., 13
Vocabularius Ex-quo, G. ten Raem, 8
Voragine, J. de. See Jacobus
Vouillième, E., 20
Vulgaria Terentii. See Terentius

W. G., binder, 64
Wake, Gerard, 89
Wakefield, Dr, 88, 89
Waltwnem, J., 98
Wandsforth, F., 43-45, 47, 49-51, 63, 64
Wandsforth, G., 44, 45, 47-49, 64
Ward, John, 89
Warwyke, E., 48
Warwyke, J., 48
Waterson, R., 47
Watson, H., 47, 53, 54
Weale, W. H. J., 95
Welles, John, 63
Wesel, printing, 108, 109
West, Nicholas, 79
Westminster Abbey Library, 12, 73, 83, 95
Weywick, M., 47, 48, 50.
See also Warwyke
Whitinton, R., Grammar, 29;
U. Mylner, 57;
J Scolar, 69;
W. de Worde, 77
Winchester binding, 30, 32
Winchester, printing, 120
Wodhull, M., 36
Wolsey, T., 76, 96, 105
Wood, Anthony à, 29
Worcester Cathedral Library, 65
Worde, W. de, 26, 29, 30, 34, 38, 41, 47, 48, 53, 54, 57, 58, 67,
69, 70, 77
Wordsworth, C., and F. Procter, Breviary, 56
Wraghton, W. See Turner
Wutton, Mr, 25
Wyer, Robert, 54
Wynkyn Lane, 29
York Minster Library, 36, 52, 64
York, Stationers’ Company, 63

Zurich, printing, 121

Zwingli, U. Certain precepts, A Scoloker, 106, 107;
Short pathway to the Scriptures, J. Oswen, 113

PRINTED BY
TURNBULL AND SPEARS
EDINBURGH

Transcriber’s Notes:
New original cover art included with this eBook is granted to the public
domain.
Variations in spelling and hyphenation are retained.
Perceived typographical errors have been changed.
 *** END OF THE PROJECT GUTENBERG EBOOK THE ENGLISH
PROVINCIAL PRINTERS, STATIONERS AND BOOKBINDERS TO 1557
***

Updated editions will replace the previous one—the old editions will
be renamed.

Creating the works from print editions not protected by U.S.

copyright law means that no one owns a United States copyright in
these works, so the Foundation (and you!) can copy and distribute it
in the United States without permission and without paying
copyright royalties. Special rules, set forth in the General Terms of
Use part of this license, apply to copying and distributing Project
Gutenberg™ electronic works to protect the PROJECT GUTENBERG™
concept and trademark. Project Gutenberg is a registered trademark,
and may not be used if you charge for an eBook, except by following
the terms of the trademark license, including paying royalties for use
of the Project Gutenberg trademark. If you do not charge anything
for copies of this eBook, complying with the trademark license is
very easy. You may use this eBook for nearly any purpose such as
creation of derivative works, reports, performances and research.
Project Gutenberg eBooks may be modified and printed and given
away—you may do practically ANYTHING in the United States with
eBooks not protected by U.S. copyright law. Redistribution is subject
to the trademark license, especially commercial redistribution.

START: FULL LICENSE

THE FULL PROJECT GUTENBERG LICENSE
 PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK

To protect the Project Gutenberg™ mission of promoting the free

distribution of electronic works, by using or distributing this work (or
any other work associated in any way with the phrase “Project
Gutenberg”), you agree to comply with all the terms of the Full
Project Gutenberg™ License available with this file or online at
www.gutenberg.org/license.

Section 1. General Terms of Use and

Redistributing Project Gutenberg™
electronic works
1.A. By reading or using any part of this Project Gutenberg™
electronic work, you indicate that you have read, understand, agree
to and accept all the terms of this license and intellectual property
(trademark/copyright) agreement. If you do not agree to abide by all
the terms of this agreement, you must cease using and return or
destroy all copies of Project Gutenberg™ electronic works in your
possession. If you paid a fee for obtaining a copy of or access to a
Project Gutenberg™ electronic work and you do not agree to be
bound by the terms of this agreement, you may obtain a refund
from the person or entity to whom you paid the fee as set forth in
paragraph 1.E.8.

1.B. “Project Gutenberg” is a registered trademark. It may only be

used on or associated in any way with an electronic work by people
who agree to be bound by the terms of this agreement. There are a
few things that you can do with most Project Gutenberg™ electronic
works even without complying with the full terms of this agreement.
See paragraph 1.C below. There are a lot of things you can do with
Project Gutenberg™ electronic works if you follow the terms of this
agreement and help preserve free future access to Project
Gutenberg™ electronic works. See paragraph 1.E below.
1.C. The Project Gutenberg Literary Archive Foundation (“the
Foundation” or PGLAF), owns a compilation copyright in the
collection of Project Gutenberg™ electronic works. Nearly all the
individual works in the collection are in the public domain in the
United States. If an individual work is unprotected by copyright law
in the United States and you are located in the United States, we do
not claim a right to prevent you from copying, distributing,
performing, displaying or creating derivative works based on the
work as long as all references to Project Gutenberg are removed. Of
course, we hope that you will support the Project Gutenberg™
mission of promoting free access to electronic works by freely
sharing Project Gutenberg™ works in compliance with the terms of
this agreement for keeping the Project Gutenberg™ name associated
with the work. You can easily comply with the terms of this
agreement by keeping this work in the same format with its attached
full Project Gutenberg™ License when you share it without charge
with others.

1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside the
United States, check the laws of your country in addition to the
terms of this agreement before downloading, copying, displaying,
performing, distributing or creating derivative works based on this
work or any other Project Gutenberg™ work. The Foundation makes
no representations concerning the copyright status of any work in
any country other than the United States.

1.E. Unless you have removed all references to Project Gutenberg:

1.E.1. The following sentence, with active links to, or other

immediate access to, the full Project Gutenberg™ License must
appear prominently whenever any copy of a Project Gutenberg™
work (any work on which the phrase “Project Gutenberg” appears,
or with which the phrase “Project Gutenberg” is associated) is
accessed, displayed, performed, viewed, copied or distributed:
 This eBook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this eBook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.

1.E.2. If an individual Project Gutenberg™ electronic work is derived

from texts not protected by U.S. copyright law (does not contain a
notice indicating that it is posted with permission of the copyright
holder), the work can be copied and distributed to anyone in the
United States without paying any fees or charges. If you are
redistributing or providing access to a work with the phrase “Project
Gutenberg” associated with or appearing on the work, you must
comply either with the requirements of paragraphs 1.E.1 through
1.E.7 or obtain permission for the use of the work and the Project
Gutenberg™ trademark as set forth in paragraphs 1.E.8 or 1.E.9.

1.E.3. If an individual Project Gutenberg™ electronic work is posted

with the permission of the copyright holder, your use and distribution
must comply with both paragraphs 1.E.1 through 1.E.7 and any
additional terms imposed by the copyright holder. Additional terms
will be linked to the Project Gutenberg™ License for all works posted
with the permission of the copyright holder found at the beginning
of this work.

1.E.4. Do not unlink or detach or remove the full Project

Gutenberg™ License terms from this work, or any files containing a
part of this work or any other work associated with Project
Gutenberg™.

1.E.5. Do not copy, display, perform, distribute or redistribute this

electronic work, or any part of this electronic work, without
prominently displaying the sentence set forth in paragraph 1.E.1
with active links or immediate access to the full terms of the Project
Gutenberg™ License.

1.E.6. You may convert to and distribute this work in any binary,
compressed, marked up, nonproprietary or proprietary form,
including any word processing or hypertext form. However, if you
provide access to or distribute copies of a Project Gutenberg™ work
in a format other than “Plain Vanilla ASCII” or other format used in
the official version posted on the official Project Gutenberg™ website
(www.gutenberg.org), you must, at no additional cost, fee or
expense to the user, provide a copy, a means of exporting a copy, or
a means of obtaining a copy upon request, of the work in its original
“Plain Vanilla ASCII” or other form. Any alternate format must
include the full Project Gutenberg™ License as specified in
paragraph 1.E.1.

1.E.7. Do not charge a fee for access to, viewing, displaying,

performing, copying or distributing any Project Gutenberg™ works
unless you comply with paragraph 1.E.8 or 1.E.9.

1.E.8. You may charge a reasonable fee for copies of or providing

access to or distributing Project Gutenberg™ electronic works
provided that:

• You pay a royalty fee of 20% of the gross profits you derive
from the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
 about donations to the Project Gutenberg Literary Archive
Foundation.”

• You provide a full refund of any money paid by a user who

notifies you in writing (or by e-mail) within 30 days of receipt
that s/he does not agree to the terms of the full Project
Gutenberg™ License. You must require such a user to return or
destroy all copies of the works possessed in a physical medium
and discontinue all use of and all access to other copies of
Project Gutenberg™ works.

• You provide, in accordance with paragraph 1.F.3, a full refund of

any money paid for a work or a replacement copy, if a defect in
the electronic work is discovered and reported to you within 90
days of receipt of the work.

• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.

1.E.9. If you wish to charge a fee or distribute a Project Gutenberg™

electronic work or group of works on different terms than are set
forth in this agreement, you must obtain permission in writing from
the Project Gutenberg Literary Archive Foundation, the manager of
the Project Gutenberg™ trademark. Contact the Foundation as set
forth in Section 3 below.

1.F.

1.F.1. Project Gutenberg volunteers and employees expend

considerable effort to identify, do copyright research on, transcribe
and proofread works not protected by U.S. copyright law in creating
the Project Gutenberg™ collection. Despite these efforts, Project
Gutenberg™ electronic works, and the medium on which they may
be stored, may contain “Defects,” such as, but not limited to,
incomplete, inaccurate or corrupt data, transcription errors, a
copyright or other intellectual property infringement, a defective or
damaged disk or other medium, a computer virus, or computer
codes that damage or cannot be read by your equipment.

1.F.2. LIMITED WARRANTY, DISCLAIMER OF DAMAGES - Except for

the “Right of Replacement or Refund” described in paragraph 1.F.3,
the Project Gutenberg Literary Archive Foundation, the owner of the
Project Gutenberg™ trademark, and any other party distributing a
Project Gutenberg™ electronic work under this agreement, disclaim
all liability to you for damages, costs and expenses, including legal
fees. YOU AGREE THAT YOU HAVE NO REMEDIES FOR
NEGLIGENCE, STRICT LIABILITY, BREACH OF WARRANTY OR
BREACH OF CONTRACT EXCEPT THOSE PROVIDED IN PARAGRAPH
1.F.3. YOU AGREE THAT THE FOUNDATION, THE TRADEMARK
OWNER, AND ANY DISTRIBUTOR UNDER THIS AGREEMENT WILL
NOT BE LIABLE TO YOU FOR ACTUAL, DIRECT, INDIRECT,
CONSEQUENTIAL, PUNITIVE OR INCIDENTAL DAMAGES EVEN IF
YOU GIVE NOTICE OF THE POSSIBILITY OF SUCH DAMAGE.

1.F.3. LIMITED RIGHT OF REPLACEMENT OR REFUND - If you

discover a defect in this electronic work within 90 days of receiving
it, you can receive a refund of the money (if any) you paid for it by
sending a written explanation to the person you received the work
from. If you received the work on a physical medium, you must
return the medium with your written explanation. The person or
entity that provided you with the defective work may elect to provide
a replacement copy in lieu of a refund. If you received the work
electronically, the person or entity providing it to you may choose to
give you a second opportunity to receive the work electronically in
lieu of a refund. If the second copy is also defective, you may
demand a refund in writing without further opportunities to fix the
problem.

1.F.4. Except for the limited right of replacement or refund set forth
in paragraph 1.F.3, this work is provided to you ‘AS-IS’, WITH NO
OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.

1.F.5. Some states do not allow disclaimers of certain implied

warranties or the exclusion or limitation of certain types of damages.
If any disclaimer or limitation set forth in this agreement violates the
law of the state applicable to this agreement, the agreement shall be
interpreted to make the maximum disclaimer or limitation permitted
by the applicable state law. The invalidity or unenforceability of any
provision of this agreement shall not void the remaining provisions.

1.F.6. INDEMNITY - You agree to indemnify and hold the Foundation,

the trademark owner, any agent or employee of the Foundation,
anyone providing copies of Project Gutenberg™ electronic works in
accordance with this agreement, and any volunteers associated with
the production, promotion and distribution of Project Gutenberg™
electronic works, harmless from all liability, costs and expenses,
including legal fees, that arise directly or indirectly from any of the
following which you do or cause to occur: (a) distribution of this or
any Project Gutenberg™ work, (b) alteration, modification, or
additions or deletions to any Project Gutenberg™ work, and (c) any
Defect you cause.

Section 2. Information about the Mission

of Project Gutenberg™
Project Gutenberg™ is synonymous with the free distribution of
electronic works in formats readable by the widest variety of
computers including obsolete, old, middle-aged and new computers.
It exists because of the efforts of hundreds of volunteers and
donations from people in all walks of life.

Volunteers and financial support to provide volunteers with the

assistance they need are critical to reaching Project Gutenberg™’s
goals and ensuring that the Project Gutenberg™ collection will
remain freely available for generations to come. In 2001, the Project
Gutenberg Literary Archive Foundation was created to provide a
secure and permanent future for Project Gutenberg™ and future
generations. To learn more about the Project Gutenberg Literary
Archive Foundation and how your efforts and donations can help,
see Sections 3 and 4 and the Foundation information page at
www.gutenberg.org.

Section 3. Information about the Project

Gutenberg Literary Archive Foundation
The Project Gutenberg Literary Archive Foundation is a non-profit
501(c)(3) educational corporation organized under the laws of the
state of Mississippi and granted tax exempt status by the Internal
Revenue Service. The Foundation’s EIN or federal tax identification
number is 64-6221541. Contributions to the Project Gutenberg
Literary Archive Foundation are tax deductible to the full extent
permitted by U.S. federal laws and your state’s laws.

The Foundation’s business office is located at 809 North 1500 West,

Salt Lake City, UT 84116, (801) 596-1887. Email contact links and up
to date contact information can be found at the Foundation’s website
and official page at www.gutenberg.org/contact

Section 4. Information about Donations to

the Project Gutenberg Literary Archive
Foundation
Project Gutenberg™ depends upon and cannot survive without
widespread public support and donations to carry out its mission of
increasing the number of public domain and licensed works that can
be freely distributed in machine-readable form accessible by the
widest array of equipment including outdated equipment. Many
small donations ($1 to $5,000) are particularly important to
maintaining tax exempt status with the IRS.

The Foundation is committed to complying with the laws regulating

charities and charitable donations in all 50 states of the United
States. Compliance requirements are not uniform and it takes a
considerable effort, much paperwork and many fees to meet and
keep up with these requirements. We do not solicit donations in
locations where we have not received written confirmation of
compliance. To SEND DONATIONS or determine the status of
compliance for any particular state visit www.gutenberg.org/donate.

While we cannot and do not solicit contributions from states where

we have not met the solicitation requirements, we know of no
prohibition against accepting unsolicited donations from donors in
such states who approach us with offers to donate.

International donations are gratefully accepted, but we cannot make

any statements concerning tax treatment of donations received from
outside the United States. U.S. laws alone swamp our small staff.

Please check the Project Gutenberg web pages for current donation
methods and addresses. Donations are accepted in a number of
other ways including checks, online payments and credit card
donations. To donate, please visit: www.gutenberg.org/donate.

Section 5. General Information About

Project Gutenberg™ electronic works
Professor Michael S. Hart was the originator of the Project
Gutenberg™ concept of a library of electronic works that could be
freely shared with anyone. For forty years, he produced and
distributed Project Gutenberg™ eBooks with only a loose network of
volunteer support.
Project Gutenberg™ eBooks are often created from several printed
editions, all of which are confirmed as not protected by copyright in
the U.S. unless a copyright notice is included. Thus, we do not
necessarily keep eBooks in compliance with any particular paper
edition.

Most people start at our website which has the main PG search
facility: www.gutenberg.org.

This website includes information about Project Gutenberg™,

including how to make donations to the Project Gutenberg Literary
Archive Foundation, how to help produce our new eBooks, and how
to subscribe to our email newsletter to hear about new eBooks.
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!

ebooknice.com