所需积分/C币:9 2019-09-12 07:56:08 7.03MB PDF

NEURAL DATA SCIENCE A Primer with matlab and python Tm ERIK LEE NYLEN Parsec media. New york NY, United states PASCAL WALLISCH New york university, New york Ny, United states ACADEMIC PRESS An imprint of elsevier ELSEVIER elsevier. com Academic Press is an imprint of elsevier compounds, or experiments described herein. In using such information 125 London Wall, London eC2Y 5AS, United Kingdom or methods they should be mindful of their own safety and the 525 B Street, Suite 1800, San Diego, Ca92101-4495, United States safety of others, including parties for whom they have a professional 50 Hampshire Street, 5th Floor, Cambridge, MA02139, United States responsibility The Boulevard, langford Lane Kidlington, Oxford OX5 1GB To the fullest extent of the law neither the publisher nor the authors United Kingdom contributors, or editors, assume any liability for any injury and/ Copyright o 2017 Elsevier Inc. All rights reserved or damage to persons or property as a matter of products liability, gligence or otherwise, or from any use or operation of any methods, MATLAB is a trademark of The math Works, inc. and is used with products, instructions, or ideas contained in the material herein permission British Library Cataloguing-in-Publication Data The Math Works does not warrant the accuracy of the text or exercises in A catalogue record for this book is available from the British Library this book Library of Congress Cataloging-in-Publication Data This book's use or discussion of MATLAB software or related products A catalog record for this book is available from the Library of Congress does not constitute endorsement or sponsorship by The Math Works of ISBN:978-0-12-804043-0 a particular pedagogical approach or particular use of the MatLaB software For Information on all academic Press publications PythonTM is a trademark of Python Software Foundation visitourwebsiteat 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 Working together permission in writing from the publisher. Details on how to seek permission, further information about the Publisher's permissions to grow libraries in policies and our arrangements with organizations such as the Copyright ELSE Book Aid International developing countries Clearance Center and the Copyright Licensing Agency, can be found at This book and the individual contributions contained in it are protected under copyright by the Publisher(other than as may be noted herein Publisher: Mara Conner Notices Acquisition Editor: Natalie Farra Knowledge and best practice in this field are constantly changing. As new Editorial Project Manager: Kathy Padilla research and experience broaden our understanding, changes in research Production Project Manager: Edward Taylor methods, professional practices, or medical treatment may become Designer: Mark rogers necessary Typeset by MPS Limited, Chennai, India Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, Dedication Dedicated to our grandfathers Karl radle and lee nyle Biography Erik Lee nylen received his Phd from the Center for mathematical tools in neuroscience and Neural Science at New York University, and his Bse and ms He received his PhD in Psychology from the University in Biomedical Engineering at the University of lowa. He did of Chicago and worked as a postdoctoral fellow at the a fellowship at Insight Data Science, and has taught at the Center for Neural Science at New York University. He Neural data Science summer course at Cold spring harbor has a long-term commitment and is dedicated to educa Laboratory. He is a patented inventor and has performed tional excellence, which was recognized by the"Wayne with numerous musical groups. He is currently a data sci- C. Booth Graduate Student Prize for Excellence in teach- entist in New York, where he also is Executive Codirector of ing"at the University of Chicago and the "Golden Dozen The Stand, the New York city dance marathon Award"at New York University. He cofounded and coor Pascal Wallisch serves as a professor in the Department ganizes the"Neural Data Science"summer course at col of Psychology at New York University where he cur- Spring Harbor Laboratory and coauthored Matlab for rently teaches statistics, programming, and the use of Neuroscientists P1 reface The future of neuroscience is likely to be very differ- Kording, 2011). Over time, a continuously rising share o ent from its past. Even though the field of neuroscience the latent information inherent in ongoing neural pro- is inherently interdisciplinary, it has been dominated by cesses will become available in the form of data to inform neurobiology, a subfield of biology. As a corollary, the field our creation of models and theories of how the brain has been in a state of relative data scarcity up to this point. works. It is critical that this development is met by a large However, biology itself is becoming ever more quantita- cadre of neuroscientists who are ready and able to handle tive, and so is neuroscience. This development is paral- these big data leled by an ever-increasing torrent of brain-derived data That is where this book comes in . Data Science, an due to the availability of large-scale data recording tech- emerging field that is dedicated to the understanding niques. Put simply, neuroscientists of the future will rou- of patterns in large datasets can, and we believe should tinely be expected to handle massive amounts of data. be brought to bear on neural data. Therefore, we aim to Given the current state of the field, we expect this transi- introduce "Neural Data Science"principles from data tion to be challenging, as the ability to do so is not within science applied to neural data(with all its inherent com the standard repertoire of neuroscience training But it also plexities and idiosyncrasies) to a broader audience of offers tremendous opportunities--it has been pointed out neuroscientists that available neuroscience methods are in principle insuf- Thus, this book is aimed at people who want to get a ficient to understand the basic workings of even simple start in this field, as we believe they will need to in order information-processing systems where ground truth is for continued professional success within the neurosci known (onas Kording, 2016). The brain is vastly more ence of the future. In order to reach as broad an audience complicated, perhaps not even primarily an information- as possible, we deliberately do not presuppose anything,no processing device and came about as a result of complex prior experience with scientific programming, linear alge- evolutionary processes with unknown design principles. bra, calculus, machine learning, or statistics. This approach To have any hope of understanding this extremely com- (conceptualized as"0 to 1 teaching")is baked in through plex organ(and the mind it gives rise to) neuroscience will out the book. All that the prospective reader(you, as you have to change. Fortunately, our ability to record more and are reading this) is expected to possess is a burning desire more data from the brain and the physiological processes to learn how to handle"big data"in neuroscience. On the going on within it is ever-increasing, as are our methods flip-side, this book is explicitly not aimed at people who to analyze these data, albeit with some lag(Stevenson are already well versed in these topics. The stated purpose XIV PREFACE of this book is to build solid foundations-to introduce the on the forefront of knowledge are mostly concerned about most relevant concepts in as clear a fashion as possible--so leaves and might even lose an appreciation for founda- that the reader can build whatever structure they wish on tional knowledge. In contrast, this is a book about roots and these foundations. We cannot anticipate specific idiosyn- trunks--introducing this foundational knowledge and how cratic use cases nor where the field itself will be going in it came about, although well also explore some branches the long term but the structure of knowledge works to our and leaves that are of particular interest or exemplary. We advantage: In general, knowledge tends to be organized believe that by the end of the book, you will be in a perfect in a fractal or tree-like structure(see cartoon below). a tree position to go from 1 to 100, or beyond, whatever that might has roots, a trunk, branches, and leaves. Experts working be for your particular use case, with confidence Specifically, this book is divided into four parts: In the neuroscience"from a modern perspective. This is impor first part, well explore philosophical underpinnings of tant because this classical approach still dominates neu- the field and conceptual foundations that are truly ele- roscience today and you would be well advised to be mentary. In the second part, we cover techniques from a familiar with it before moving on to the next level. The field that could properly be described as"computational next level(Part 3)consists of the burgeoning data science PREFACE techniques applied to neural data. The final part of the Giacomucci, Mark Rogers and Rob duckwall at our pub book comprises several appendices that are self-con- lisher for their understanding, support, and patience. We tained repositories of useful information that didn't fit would like to thank Julie Cachia for her helpful comments any of the other parts on the manuscript. We are grateful to Matthew A. Smith One design decision we made concerns the fact that we for sharing neural data as well as to Mike X Cohen and are covering all material(wherever possible) in a prac- Mark Reimers for helpful feedback on an earlier version of tical fashion, with (commented) executable code in the this manuscript. We also would like to thank everyone else two most relevant languages of scientific computing- who has helped us along the way but whom we did not MATLAB(still the most prevalent language in neurosci- mention explicitly. Finally, we thank you-the reader--for ence)and Python(the most prevalent language in data the trust you place in us for considering to work through science and becoming ever more relevant in neuroscience). this book. We hope that we provide sufficient value to you In addition to this code, each chapter contains an exposi- in return. We would also like to stress that all remaining tion of what we are trying to accomplish-the strategic errors and shortcomings of this book are ours, and ours goal, why it is important, a discussion of several tactical alone approaches (algorithms) to achieve this goal, and practi- Neuroscience is undergoing radical, transformational cal considerations that arise as we implement these algo- change. We believe that the only way for neuroscience rithms in code practitioners to prosper long term is if they embrace these Writing a helpful book(as we hope this one will be)is developments, as they open up new and exciting ways to a nontrivial, yet delightful, challenge and fundamentally a understand the brain. We hope that this book will serve as team effort. Thus, we are indebted to a bevy of individuals a useful guide for all those who are in the process of mak- without whom we could not have hoped to succeed. First ing this journey and foremost, we would like to emphasize the unwaver Pascal wallisch ng support we received from our respective families and Erik lee nylen friends. Second, we are indebted to Kathy Padilla, natalie New York, june 2016 Farra, Mica Haley, and edward taylor, calum ross Karen How to Use This book as the content of this book is somewhat technical in The pseudocode portion will have a dedicated panel,next nature, we want to be explicit about how to use it, so that to a panel that contains a figure resulting from the corre- you will be able to get the most out of it sponding code(either Matlab or Python), where appro What is this book trying to do and who is it aimed at? The priate. This approach lets the reader learn Python from point of this book is to make the reader code-safe. This Matlab, Matlab from Python, Python from English, Matlab means that if you work through this book (either in the from English, or even learn English from prior knowledge context of a course or by yourself), we believe that you of either Python or Matlab will be able to explore any topic within the scope of neu ral data science you desire in more detail, without drown- Python Matlab Pseudocode Figure ing. Actual programming proficiency will take thousands >>>5+7 >>5+7 Here we simply No relevant figure of hours of deliberate engagement with some problem ans=12 add five and seven here area, actively coding. There are no good shortcuts to this and the output but the point of this book is to put in a couple of hundred twelve is printed carefully structured(and guided) hours so that you can explore with confidence, later Sometimes, where appropriate due to the spatial The Rosetta Stone Approach. We use as inspiration the arrangements of the data in question, we will use a hori original Rosetta stone (Champollion, 1828). To make zontal arrangement, like in the original Rosetta stone that the book accessible to the largest possible audience, we dates from 196 BC decided to cover all materials in the most relevant lan guages in parallel, wherever possible. We recognize that Pseudocode Sum up the numbers in the vector the native language of most current neuroscientists will be Matlab, and we want to enable them to readily switch Python In>>>sum([0,0,0,0,0,0,0,0,0,1,0,1,0,1 0,0,0,1,0,0,0]) back and forth between Python and Matlab, should they Out >>>4 desire to do so. Our language of choice for pseudocode--tneMATLAB words that describe what the code is supposed to do-is >>sum([0,0,0,0,0,0,0,0,0,1,0,1,0,1,0,0 ,0,1,0,0,0]) English. We also use English for the prose that specifies ans 4 the strategic goals and the algorithms to implement them XVIl

  • 至尊王者


关注 私信 TA的资源