• PDF / 4,015,010 Bytes
• 191 Pages / 439.371 x 683.151 pts Page_size
• 56 Downloads / 201 Views

DOWNLOAD

REPORT

---

Dual Learning

Dual Learning

Tao Qin

Dual Learning

Tao Qin Microsoft Research Asia (China) Beijing, China

ISBN 978-981-15-8883-9 ISBN 978-981-15-8884-6 (eBook) https://doi.org/10.1007/978-981-15-8884-6 © Springer Nature Singapore Pte Ltd. 2020 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, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

I would like to dedicate this book to my wife and my lovely son!

Preface

Deep neural networks have become the dominant paradigm for artificial intelligence (AI) in the past decade, and deep learning has significantly advanced various areas of AI, spanning from computer vision, natural language and speech, to game playing. A key factor contributing to the success of deep learning is availability of large-scale labeled training data. Correspondingly, a leading challenge (and also a hot research direction) for deep learning is how to learn from limited/insufficient labeled data. Dual learning is a new learning framework proposed to address this challenge by leveraging structural duality between AI tasks. This book gives a comprehensive review of recent research on dual learning. We introduce its basic principles, including dual reconstruction, joint-probability equation, and marginal probability equation, and cover various learning settings and algorithms, including dual semi-supervised learning, dual unsupervised learning, dual supervised learning, and dual inference. For each setting, we introduce diverse applications, such as machine translation, image-to-image translation, speech synthesis and recognition, question answering and generation, image classification and generation, code summarization and generation, sentiment analysis, etc. This book is written for re