DOCSLIB.ORG

Fine-Tuning a Pre-Trained Comment-Domain BERT Model

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Fine-Tuning a Pre-Trained Comment-Domain BERT Model Jigsaw @ AMI and HaSpeeDe2: Fine-Tuning a Pre-Trained Comment-Domain BERT Model Alyssa Lees and Jeffrey Sorensen and Ian Kivlichan Google Jigsaw New York, NY (alyssalees|sorenj|kivlichan)@google.com Abstract labeled by crowd workers. Note that Perspec- tiveAPI actually hosts a number of different mod- The Google Jigsaw team produced els that each score different attributes. The under- submissions for two of the EVALITA lying technology and performance of these models 2020 (Basile et al., 2020) shared tasks, has evolved over time. based in part on the technology that pow- While Jigsaw has hosted three separate Kaggle ers the publicly available PerspectiveAPI competitions relevant to this competition (Jigsaw, comment evaluation service. We present a 2018; Jigsaw, 2019; Jigsaw, 2020) we have not basic description of our submitted results traditionally participated in academic evaluations. and a review of the types of errors that our system made in these shared tasks. 3 Related Work The models we build are based on the popular 1 Introduction BERT architecture (Devlin et al., 2019) with dif- The HaSpeeDe2 shared task consists of Italian so- ferent pre-training and fine-tuning approaches. cial media posts that have been labeled for hate In part, our submissions explore the importance speech and stereotypes. As Jigsaw’s participation of pre-training (Gururangan et al., 2020) in the was limited to the A and B tasks, we will be lim- context of toxicity and the various competition at- iting our analysis to that portion. The full details tributes. A core question is to what extent these of the dataset are available in the task guidelines domains overlap. Jigsaw’s customized models (Bosco et al., 2020). (used for the second HaSpeeDe2 submission, and The AMI task includes both raw (natural Twit- both AMI submissions) are pretrained on a set of ter) and synthetic (template-generated) datasets. one billion user-generated comments: this imparts The raw data consists of Italian tweets manually statistical information to the model about com- labelled and balanced according to misogyny and ments and conversations online. This model is fur- aggressiveness labels, while the synthetic data is ther fine-tuned on various toxicity attributes (toxi- labelled only for misogyny and is intended to city, severe toxicity, profanity, insults, identity at- measure the presence of unintended bias (Elisa- tacks, and threats), but it is unclear how well these betta Fersini, 2020). should align with the competition attributes. The descriptions of these attributes and how they were 2 Background collected from crowd workers can be found in the data descriptions for the Jigsaw Unintended Bias Jigsaw, a team within Google, develops the Per- in Toxicity Classification (Jigsaw, 2019) website. spectiveAPI machine learning comment scoring A second question studied in prior work is to system, which is used by numerous social media what extent training generalizes across languages companies and publishers. Our system is based (Pires et al., 2019; Wu and Dredze, 2019; Pa- on distillation and uses a convolutional neural- mungkas et al., 2020). The majority of our train- network to score individual comments according ing data is English comment data from a variety to several attributes using supervised training data of sources, while this competition is based on Ital- ian Twitter data. Though multilingual transfer has Copyright ©2020 for this paper by its authors. Use per- mitted under Creative Commons License Attribution 4.0 In- been studied in general contexts, less is known ternational (CC BY 4.0). about the specific cases of toxicity, hate speech, misogyny, and harassment. This was one of the fo- gual teacher model (that is too large to practi- cuses of Jigsaw’s recent Kaggle competition (Jig- cally serve in production) to a smaller CNN. Using saw, 2020); i.e., what forms of toxicity are shared this large teacher model, we initially compared the across languages (and hence can be learned by EVALITA hate speech and stereotype annotations multilingual models) and what forms are different. against the teacher model’s scores for different at- tributes. The results are shown in Figure 1 for the 4 Submission Details training data. Perspective is a reasonable detector for the hate speech attribute, but performs less well Attribute hs Attribute stereotype 1.0 for the stereotype attribute, with the identity attack 0.8 model performing the best. Using these same models on the AMI task, 0.6 shown in Figure 2 for detecting misogyny proved 0.4 even more challenging. In this case, the aggres- True Positive Rate identity_attack 83.6% identity_attack 71.2% severe_toxicity 82.6% severe_toxicity 70.9% siveness attribute was evaluated only on the sub- 0.2 toxicity 82% insult 70.5% insult 80.8% toxicity 70.4% profanity 79% profanity 69.9% set of the training data labeled misogynous. In threat 68.1% threat 63.6% 0.0 this case, the most popular attribute of “toxicity” 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate False Positive Rate is actually counter-indicative of the misogyny la- bel. The best detector for both of these attributes Figure 1: ROC curves for the PerspectiveAPI appears to be the “threat” model. multilingual teacher model attributes compared to As can be seen, the existing classifiers are all the HaSpeeDe2 attributes (hate speech and stereo- poor predictors of both attributes for this shared type). task. Due to errors in our initial analysis, we did not end up using any of the models used for Per- Attribute misogynous Attribute aggressiveness spectiveAPI in our final submissions. 1.0 1.0 0.8 0.8 0.6 0.6 Category Submissionhate speechstereotype 0.4 0.4 news 1 0.68 0.64 True Positive Rate threat 61.5% threat 66% identity_attack 56.5% identity_attack 61.8% 2 0.64 0.68 0.2 0.2 insult 45.2% severe_toxicity 53.9% severe_toxicity 44.4% insult 53.3% tweets 1 0.72 0.67 toxicity 39.8% toxicity 48.4% profanity 28.8% 0.0 profanity 37% 2 0.77 0.74 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate False Positive Rate Table 1: Macro-averaged F1 scores for Jigsaw’s HaSpeeDe2 Submissions. Figure 2: ROC curves for PerspectiveAPI multi- lingual teacher model attributes compared to the AMI attributes (misogyny and aggressiveness). 4.1 HaSpeeDe2 The Jigsaw team submitted two separate submis- As Jigsaw has already developed toxicity mod- sions that were independently trained for Tasks A els for the Italian language, we initially hoped and B. that these would provide a preliminary baseline for the competition despite the independent na- 4.1.1 First Submission ture of the development of the annotation guide- Our first submission, one that did not perform very lines. Our Italian models score comments for tox- well, was based on a simple multilingual BERT icity as well as five additional distinct toxicity at- model fine-tuned on 10 random splits of the train- tributes: severe toxicity, profanity, threats, insults, ing data. For each split, 10% of the data was and identity attacks. We might expect some of held out to choose an appropriate equal-error-rate these attributes to correlate with the HaSpeeDe2 threshold for the resulting model. and AMI attributes, though it is not immediately The BERT fine-tuning system used the 12 layer clear whether any of these correlations should be model (Tensorflow Hub, 2020), a batch size of particularly strong. 64 and sequence length of 128. A single dense The current Jigsaw PerspectiveAPI models are layer is used to connect to the two output sigmoids typically trained via distillation from a multilin- which are trained using a binary cross-entropy loss using stochastic gradient descent with early stop- ping, which is computed using the AUC metric 1.0 computed using the 10% held out slice. This 0.8 model is implemented using Keras (Chollet and others, 2015). 0.6 To create the final submission, the decisions of the ten separate classifiers were combined in a ma- 0.4 True Positive Rate jority voting scheme (if 5 or more models pro- Tweets Hatespeech 85.5% 0.2 Tweets Stereotype 82.6% duced a positive detection, the attribute was as- News Stereotype 77.3% signed true). News Hatespeech 75.7% 0.0 0.0 0.2 0.4 0.6 0.8 1.0 4.1.2 Second Submission False Positive Rate Our second submission was based on a similar ap- proach of fine-tuning a BERT-based model, but Figure 3: ROC plots for HaSpeeDe2 Test Set La- one based on a more closely matched training set. bels. The underlying technology we used is the same as the Google Cloud AutoML for natural language point and custom wordpiece vocabulary from Sec- processing product that had been employed in sim- tion 4.1.2. However, a larger batch-size of 128 ilar labeling applications (Bisong, 2019). was specified. All models were fine-tuned simul- The remaining models built for this competi- taneously on misogynous and aggressive labels tion and in the subsequent section are based on a using the provided data, where zero aggressive- customized BERT 768-dimension 12-layer model ness weights were assigned to data points with no pretrained on 1B user-generated comments using misogynous labels. MLM for 125 steps. This model was then fine- tuned on supervised comments in multiple lan- Both submissions were based on ensembles of guages for six attributes: toxicity, severe toxic- partitioned models evaluated on a 10% held-out ity, obscene, threat, insult, and identity hate.
Load more

Recommended publications
  • GOOGLE LLC V. ORACLE AMERICA, INC
    (Slip Opinion) OCTOBER TERM, 2020 1 Syllabus NOTE: Where it is feasible, a syllabus (headnote) will be released, as is being done in connection with this case, at the time the opinion is issued. The syllabus constitutes no part of the opinion of the Court but has been prepared by the Reporter of Decisions for the convenience of the reader. See United States v. Detroit Timber & Lumber Co., 200 U. S. 321, 337. SUPREME COURT OF THE UNITED STATES Syllabus GOOGLE LLC v. ORACLE AMERICA, INC. CERTIORARI TO THE UNITED STATES COURT OF APPEALS FOR THE FEDERAL CIRCUIT No. 18–956. Argued October 7, 2020—Decided April 5, 2021 Oracle America, Inc., owns a copyright in Java SE, a computer platform that uses the popular Java computer programming language. In 2005, Google acquired Android and sought to build a new software platform for mobile devices. To allow the millions of programmers familiar with the Java programming language to work with its new Android plat- form, Google copied roughly 11,500 lines of code from the Java SE pro- gram. The copied lines are part of a tool called an Application Pro- gramming Interface (API). An API allows programmers to call upon prewritten computing tasks for use in their own programs. Over the course of protracted litigation, the lower courts have considered (1) whether Java SE’s owner could copyright the copied lines from the API, and (2) if so, whether Google’s copying constituted a permissible “fair use” of that material freeing Google from copyright liability. In the proceedings below, the Federal Circuit held that the copied lines are copyrightable.
    [Show full text]
  • The Aesthetic Mind This Page Intentionally Left Blank the Aesthetic Mind Philosophy and Psychology
    The Aesthetic Mind This page intentionally left blank The Aesthetic Mind Philosophy and Psychology EDITED BY Elisabeth Schellekens and Peter Goldie 1 3 Great Clarendon Street, Oxford OX26DP Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York # the several contributors 2011 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2011 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available Typeset by SPI Publisher Services, Pondicherry, India Printed in Great Britain on acid-free paper by MPG Books Group, Bodmin and King’s Lynn ISBN 978–0–19–969151–7 13579108642 Contents List of Figures viii Notes on Contributors ix Introduction 1 Elisabeth Schellekens and Peter Goldie Part I.
    [Show full text]
  • A Toolkit to Develop a Virtual Dementia Friendly Community
    An Australian Government Initiative Government An Australian Friendly Community Friendly a Virtual Dementia A Toolkit to develop develop to A Toolkit Verily Connect | An Australian Government Initiative A Toolkit to develop a Virtual Dementia Friendly Community Authors Blackberry, I., Wilding, C., Morgan, D., Winbolt, M., Greenhill, J., Perkins, D., O’Connell, M., Bauer, M., Morley, C., Farmer, J., Royals, K., Rasekaba, T., Hamiduzzaman, M., Gottschall, K., Robinson, A., Zaplin, E., & Davis, H. Year February 2020 Recommended citation Blackberry, I., Wilding, C., Morgan, D., Winbolt, M., Greenhill, J., Perkins, D., O’Connell, M., Bauer, M., Morley, C., Farmer, J., Royals, K., Rasekaba, T., Hamiduzzaman, M., Gottschall, K., Robinson, A., Zaplin, E., & Davis, H. 2020. Verily Connect Toolkit: how to develop a Virtual Dementia-Friendly Community. John Richards Centre for Rural Ageing Research, La Trobe University, Wodonga. Disclaimer The information set out in this toolkit is current at the date of this publication and is intended for use as a guide of a general nature only and may or may not be relevant to people living with dementia or their carers. Nor is this toolkit exhaustive of the subject matter. Persons implementing any advice contained in this toolkit must exercise their own skill, judgement and seek appropriate professional advice relevant to the matter. Accordingly, The John Richards Centre for Rural Ageing Research, La Trobe University and its employees and agents shall accept no liability (including without limitation liability by reason of negligence) to any users of the information contained in this toolkit for any loss or damage (consequential or otherwise), cost or expense incurred or arising by reason of any person using or relying on the information contained in this toolkit and whether caused by reason of any error, negligent act, omission or misrepresentation in the information.
    [Show full text]
  • Jigsaw Puzzles
    Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzles Mehdi Noroozi and Paolo Favaro Institute for Informatiks University of Bern fnoroozi,[email protected] Abstract. In this paper we study the problem of image representation learning without human annotation. By following the principles of self- supervision, we build a convolutional neural network (CNN) that can be trained to solve Jigsaw puzzles as a pretext task, which requires no manual labeling, and then later repurposed to solve object classification and detection. To maintain the compatibility across tasks we introduce the context-free network (CFN), a siamese-ennead CNN. The CFN takes image tiles as input and explicitly limits the receptive field (or context) of its early processing units to one tile at a time. We show that the CFN includes fewer parameters than AlexNet while preserving the same se- mantic learning capabilities. By training the CFN to solve Jigsaw puzzles, we learn both a feature mapping of object parts as well as their correct spatial arrangement. Our experimental evaluations show that the learned features capture semantically relevant content. Our proposed method for learning visual representations outperforms state of the art methods in several transfer learning benchmarks. 1 Introduction arXiv:1603.09246v3 [cs.CV] 22 Aug 2017 Visual tasks, such as object classification and detection, have been successfully approached through the supervised learning paradigm [1,11,25,36], where one uses labeled data to train a parametric model. However, as manually labeled data can be costly, unsupervised learning methods are gaining momentum. Recently, Doersch et al. [10], Wang and Gupta [39] and Agrawal et al.
    [Show full text]
  • Mitigating Jigsaw Puzzle Problem of Fingerprint Space with Extra Mile
    RTS Assisted Mobile Localization: Mitigating Jigsaw Puzzle Problem of Fingerprint Space with Extra Mile Chao Song∗yz, Jie Wuy, Li Lu∗, and Ming Liu∗z ∗School of Computer Science and Engineering, University of Electronic Science and Technology of China yDepartment of Computer and Information Sciences, Temple University zBig Data Research Center, University of Electronic Science and Technology of China Email: fchaosong, luli2009, [email protected], [email protected] Abstract—With the development of Location Based Services are associated with the locations where they are recorded. (LBSs), both academic researchers and industries have paid However, for the outdoor mobile localization, it is difficult to more attention to GPS-less mobile localization on mobile phones. record and process the fingerprints of all the positions along The majority of the existing localization approaches have uti- the trajectory, due to the large amount of fingerprints. Thus, lized signal-fingerprint as a metric for location determinations. the trajectory-fingerprint-based approach has been proposed However, one of the most challenging issues is the problem of for the mobile localization. The phones carried by the mo- uncertain fingerprints for building the fingerprint map, termed as the jigsaw puzzle problem. In this paper, for more accurate bile users autonomously and continuously record the ambient fingerprints of the mobile localization, we investigate the changes conditions, and position information as the fingerprint of the of Received Signal Strength Indication (RSSI) from the connected user’s trajectory [4], [8]. Paek et al. in [9] propose to utilize cell-towers over time along the mobile users’ trajectories, termed a cell-ID sequence-based localization, which is recorded by as RSSI Time Series (RTS).
    [Show full text]
  • Using Computer Vision to Solve Jigsaw Puzzles
    Using Computer Vision to Solve Jigsaw Puzzles Travis V. Allen Stanford University CS231A, Spring 2016 [email protected] Abstract 1. Introduction This project focuses on the creation of an end-to-end Many in the computer vision community have written “automatic” jigsaw puzzle solving program that assembles papers on solving the “problem” of jigsaw puzzles for a jigsaw puzzle using an image of the disassembled puzzle decades. Many of the former papers written cite the po- pieces. The solver does not use the picture of the assembled tential extension of this work to other more serious endeav- puzzle for matching purposes. Rather, the solver created for ors like reconstructing archaeological artifacts or fitting to- this project attempts to solve the lost the box conundrum by gether scanned objects while others simply do it because displaying the assembled puzzle to the user without the need it’s an interesting application of computer vision concepts. for the reference image. Needless to say, this program was This author falls in the latter camp. Several methods for created with an eye toward a potential smart phone applica- constructing jigsaw puzzles from images of the pieces were tion in the future. It should be noted that the puzzle solver considered from a theoretical standpoint before the comput- is created entirely in Matlab and heavily utilizes functions ing power and the high-resolution image capturing devices found in the Matlab Image Processing Toolbox. necessary to employ these methods could be fully realized. Solving jigsaw puzzles using computer vision (CV) is an More recently, many algorithms and methods tend toward attractive problem as it benefits directly from many of the disregarding piece shape as a discriminant entirely by us- advances in the field while still proving to be both challeng- ing “square pieces” and rely instead on the underlying im- ing and intellectually stimulating.
    [Show full text]
  • An Automatic Solver for Very Large Jigsaw Puzzles Using Genetic Algorithms
    Genet Program Evolvable Mach DOI 10.1007/s10710-015-9258-0 An automatic solver for very large jigsaw puzzles using genetic algorithms 1 1 1,2 Dror Sholomon • Omid E. David • Nathan S. Netanyahu Received: 8 August 2015 / Revised: 3 December 2015 Ó Springer Science+Business Media New York 2016 Abstract In this paper we propose the first effective genetic algorithm (GA)-based jigsaw puzzle solver. We introduce a novel crossover procedure that merges two ‘‘parent’’ solutions to an improved ‘‘child’’ configuration by detecting, extracting, and combining correctly assembled puzzle segments. The solver proposed exhibits state-of-the-art performance, as far as handling previously attempted puzzles more accurately and efficiently, as well puzzle sizes that have not been attempted before. The extended experimental results provided in this paper include, among others, a thorough inspection of up to 30,745-piece puzzles (compared to previous attempts on 22,755-piece puzzles), using a considerably faster concurrent implementation of the algorithm. Furthermore, we explore the impact of different phases of the novel crossover operator by experimenting with several variants of the GA. Finally, we compare different fitness functions and their effect on the overall results of the GA- based solver. Keywords Computer vision Á Genetic algorithms Á Jigsaw puzzle A preliminary version of this paper appeared in Proceedings of the IEEE Computer Vision and Pattern Recognition Conference [19]. Area Editor for Games: Moshe Sipper. & Omid E. David [email protected] Dror Sholomon [email protected] Nathan S. Netanyahu [email protected]; [email protected] 1 Department of Computer Science, Bar-Ilan University, 52900 Ramat Gan, Israel 2 Center for Automation Research, University of Maryland, College Park, MD 20742, USA 123 Genet Program Evolvable Mach 1 Introduction The problem domain of jigsaw puzzles is widely known to almost every human being from childhood.
    [Show full text]
  • Report Resumes
    REPORT RESUMES ED 020 413 VT 005 256 VOCATIONAL EDUCATION FOR RURAL AMERICA. YEARBOOK, BY- SWANSON, GORDON I. NATIONAL EDUCATION ASSN., WASHINGTON. D.C. PUB DATE 59 EDRS PRICE HF-S1.50HC NOT AVAILABLE FROM EDRS. 355P. DESCRIPTORS- RURAL AREAS, *VOCATIONAL EDUCATION, *RURAL EDUCATION; EDUCATIONAL HISTORY, EDUCATIONAL FINANCE, PROGRAM ADMINISTRATIONIAGRICULTURAL EDUCATION,BUSINESS EDUCATION, DISTRIBUTIVE EDUCATION, HOMEMAKING EDUCATION, TRADE AND INDUSTRIAL EDUCATION, INDUSTRIAL ARTS, OCCUPATIONAL GUIDANCE, EDUCATIONAL PROBLEMS! *REFERENCE A YEARBOOK ADVISORY COMMITTEE, APPOINTED BY NATIONAL EDUCATION ASSOCIATION'S DEPARTMENT OF RURAL EDUCATION AND AMERICAN. VOCATIONAL ASSOCIATION MEMBERS, APPROVED THE OUTLINE, ASSISTED IN SELECTING AUTHORS, AND REVIEWED MANUSCRIPTS FOR THIS'YEARBOOK ON VOCATIONAL EDUCATION FOR RURAL AREAS. THE DISCUSSION IS IN TERMS OF THE UNIFYING THEME THAT DESCRIBES THE PROBLEMS OF RURAL EDUCATION-THE MALDISTRIBUTION OF HUMAN, ECONOMIC, AND EDUCATIONAL RESOURCES. PART I, WHICH TREATS THE ENVIRONMENT IN WHICH VOCATIONAL EDUCATION IS OFFERED IN AMERICAN SOCIETY, COVERS ITS HISTORY, AFFECTING FORCES, PROBLEMS OF ACCOMMODATION, INSTRUCTIONAL LEVELS, FINANCE, ADMINISTRATION, AND ITS FUTURE. PART II PRESENTS THE CONTRIBUTIONS AND OPPORTUNITIES IN 1) AGRICULTURAL, BUSINESS, DISTRIBUTIVE, HOMEMAKING, TRADES AND INDUSTRIAL EDUCATION,(2) INDUSTRIAL ARTS, AND (3) VOCATIONAL GUIDANCE. THE FINAL CHAPTER EXAMINES THE CRITICAL ISSUES IN VOCATIONAL EDUCATION. OFFICIAL RECORDS AND LISTS OF OFFICERS AND MEMBERS OF THE DEPARTMENT OF RURAL EDUCATION ARE LISTED. THIS DOCUMENT IS AVAILABLE FOR $4.00 FROM DEPARTMENT OF RURAL EDUCATION, NATIONAL EDUCATION ASSOCIATION, 1201 SIXTEENTH STREET, N.W.t.WASHINGTON, D.C. 20006. (JM) PROCESS WITH MICROFICHE AND PUBLISHER'S PRICES. MICRO- FICHE REPRODUCTION ONLY. PROCESS WITHMICROFICHE PUBLISHER'S AND PRICES. MICRO- FICHE REPRODUCTIONONLY. Vocational Education for Rural America I lid Edited by GORDON I.
    [Show full text]
  • Simply-Hitchcock-1587911892. Print
    Simply Hitchcock Simply Hitchcock DAVID STERRITT SIMPLY CHARLY NEW YORK Copyright © 2017 by David Sterritt Cover Illustration by Vladymyr Lukash Cover Design by Scarlett Rugers All rights reserved. No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the publisher, except in the case of brief quotations embodied in critical reviews and certain other noncommercial uses permitted by copyright law. For permission requests, write to the publisher at the address below. [email protected] ISBN: 978-1-943657-17-9 Brought to you by http://simplycharly.com Dedicated to Mikita, Jeremy and Tanya, Craig and Kim, and Oliver, of course Contents Praise for Simply Hitchcock ix Other Great Lives xiii Series Editor's Foreword xiv Preface xv Acknowledgements xix 1. Hitch 1 2. Silents Are Golden 21 3. Talkies, Theatricality, and the Low Ebb 37 4. The Classic Thriller Sextet 49 5. Hollywood 61 6. The Fabulous 1950s 96 7. From Psycho to Family Plot 123 8. Epilogue 145 End Notes 147 Suggested Reading 164 About the Author 167 A Word from the Publisher 168 Praise for Simply Hitchcock “With his customary style and brilliance, David Sterritt neatly unpacks Hitchcock’s long career with a sympathetic but sharply observant eye. As one of the cinema’s most perceptive critics, Sterritt is uniquely qualified to write this concise and compact volume, which is the best quick overview of Hitchcock’s work to date—written with both the cineaste and the general reader in mind.
    [Show full text]
  • The Promise of Artificial Intelligence Daniel Castro and Joshua New
    The Promise of Artificial Intelligence Daniel Castro and Joshua New center for data innovation The Promise of Artificial Intelligence Daniel Castro and Joshua New October 2016 2 Table of Contents Introduction ...........................................................................................2 What Is Artificial Intelligence? .................................................................2 The Value of AI.......................................................................................3 Typology of AI Applications ......................................................................4 Examples of AI ........................................................................................7 Accessibility ......................................................................................8 Agriculture ......................................................................................11 Business Operations .........................................................................12 Consumer Convenience ....................................................................15 Disaster Prevention and Response .....................................................16 Education .......................................................................................19 Energy ............................................................................................20 Environment ...................................................................................23 Health Care: Prevention and Screening ..............................................24
    [Show full text]
  • The 3D Jigsaw Puzzle: Mapping Large Indoor Spaces
    The 3D jigsaw puzzle: mapping large indoor spaces Ricardo Martin-Brualla1, Yanling He1, Bryan C. Russell2, and Steven M. Seitz1 1University of Washington 2Intel Abstract. We introduce an approach for analyzing annotated maps of a site, together with Internet photos, to reconstruct large indoor spaces of famous tourist sites. While current 3D reconstruction algorithms often produce a set of disconnected components (3D pieces) for indoor scenes due to scene coverage or matching failures, we make use of a provided map to lay out the 3D pieces in a global coordinate system. Our approach leverages position, orientation, and shape cues extracted from the map and 3D pieces and optimizes a global objective to recover the global layout of the pieces. We introduce a novel crowd flow cue that measures how people move across the site to recover 3D geometry orientation. We show compelling results on major tourist sites. Keywords: Indoor scene reconstruction, maps, 3D jigsaw puzzle. 1 Introduction Recent breakthroughs in computer vision now allow us to model our world in 3D with extraordinary accuracy and visual fidelity from just about any set of overlapping photos [1{3]. However, a limitation of state-of-the-art 3D reconstruc- tion techniques from Internet photos is that large scenes tend to break up into a collection of disconnected pieces due to gaps in the depicted scene coverage or matching failures. Rather than a single, fully-connected Vatican model, for instance, we get a collection of smaller 3D pieces for different rooms, such as the Sistine Chapel, the Raphael Rooms, and the Hall of Maps, each having their own 3D coordinate system.
    [Show full text]
  • HAAS SOCIALLY RESPONSIBLE INVESTMENT FUND Annual
    HAAS SOCIALLY RESPONSIBLE INVESTMENT FUND Annual Report May 2018 Annual Report Haas Socially Responsible Investment Fund Table of Contents HSRIF 2017-2018 ANNUAL LETTER FROM THE FACULTY ADVISOR………………...i 1. INTRODUCTION ................................................................................................................ 1 2. PORTFOLIO PERFORMANCE ......................................................................................... 2 OVERALL PORTFOLIO........................................................................................... 2 CONCENTRATED EQUITIES PORTFOLIO ................................................................ 4 THE HSRIF’S LONG TERM PERFORMANCE .......................................................... 6 CONTRIBUTION TO THE CRB ............................................................................... 7 3. DESCRIPTION & RATIONALE FOR REBALANCING DECISIONS ............................... 8 REASSESSING THE KLD 400 ............................................................................... 8 ADDITION OF FACEBOOK ..................................................................................... 9 INCREASE IN HANES POSITION ........................................................................... 10 DECREASE IN MASTERCARD POSITION ............................................................... 10 4. DESCRIPTION & RESULTS OF STRATEGIC PROJECTS .......................................... 11 INVESTMENT BELIEFS & ESG APPROACH .......................................................... 11 IDEA
    [Show full text]