◆  Professor Bin Yu, University of California, Berkeley; Fellow of American Academy of Arts and Sciences; IEEE Fellow

Keynote topics: Three principles of data science: predictability, stability, and computability

Short Bio: Bin Yu is Chancellor’s Professor in the Departments of Statistics and of Electrical Engineering & Computer Science at the University of California at Berkeley and a former Chair of Statistics at Berkeley. She is founding co-director of the Microsoft Joint Lab at Peking University on Statistics and Information Technology. She is engaged in interdisciplinary research with scientists from genomics, neuroscience, and medicine. In order to solve data problems in these domain areas, she develops statistics and machine learning methods/algorithms and theory and integrates with domain knowledge and quantitative critical thinking in the process.
She is Member of the U.S. National Academy of Sciences and Fellow of the American Academy of Arts and Sciences. She was a Guggenheim Fellow in 2006, an Invited Speaker at ICIAM in 2011, the Tukey Memorial Lecturer of the Bernoulli Society in 2012, and 2016 Rietz Lecturer of the Institute of Mathematical Statistics (IMS). She was President of IMS in 2013-2014, and is a Fellow of IMS, ASA, AAAS and IEEE.

Abstract: In this talk, I'd like to discuss the intertwining importance and connections of three principles of data science in the title in data-driven decisions. The ultimate importance of prediction lies in the fact that future holds the unique and possibly the only purpose of all human activities, in business, education, research, and government alike.
Making prediction as its central task and embracing computation as its core, machine learning has enabled wide-ranging data-driven successes. Prediction is a useful way to check with reality. Good prediction implicitly assumes stability between past and future. Stability (relative to data and model perturbations) is also a minimum requirement for interpretability and reproducibility of data driven results. It is closely related to uncertainty assessment. Obviously, both prediction and stability principles can not be employed without feasible computational algorithms, hence the importance of computability. The three principles will be demonstrated through analytical connections, and in the context of two on-going projects, for which "data wisdom" is also indispensable. Specifically, the first project employs deep learning networks (CNNs) to understand pattern selectivities of neurons in the difficult visual cortex V4; and the second project predicts partisanship and tone of political TV ads by employing and comparing different latent variable models with a Lasso-based model.

◆  Jacob Devlin, Senior Research Scientist, MSR, USA

Keynote topics: A Practical Guide to Neural Machine Translation

Short Bio: Jacob Devlin is a Senior NLP Scientist in the Machine Translation Group at Microsoft Research in Redmond, Washington. His primary research focus is on developing fast, scalable deep learning systems for machine translation and other applications. Prior to joining Microsoft, Mr. Devlin worked as a Research Scientist at BBN Technologies in Cambridge, Massachussets. His work on machine translation received the ACL best paper award in 2014 and the NAACL best paper award in 2012. He received his B.A./M.S. in Computer Science from the University of Maryland, College Park in 2009.

Abstract: In the last two years, attentional-sequence-to-sequence neural models have become the state-of-the-art in machine translation, far surpassing the accuracy phrasal translation systems of in many scenarios.
However, these Neural Machine Translation (NMT) systems are not without their difficulties: training a model on a large-scale data set can often take weeks, and they are typically much slower at decode time than a well-optimized phrasal system. In addition, robust training of these models often relies on particular 'recipes' that are not well-explained or justified in the literature.
In the talk, I will describe a number of tricks and techniques to substantially speed up training and decoding of large-scale NMT systems. These techniques – which vary between algorithmic and engineering-focused – reduced the time required to train a large-scale NMT from two weeks to two days, and improved the decoding speed to match that of a well-optimized phrasal MT system. In addition, I will attempt to give empirical and intuitive justification for many of the choices made regarding architecture, optimization, and hyperparameters.
Although this talk will primarily focus on NMT, the techniques described here should generalize to a number of other models based on sequence-to-sequence and recurrent neural networks, such as caption generation and conversation agents.

◆  Haizhou Li, Professor, National University of Singapore, Singapore; IEEE Fellow

Keynote topics: Evaluation of Mandarin Chinese Spoken by Non-Native Speakers of European Origin

Short Bio: Haizhou Li received the B.Sc, M.Sc, and Ph.D degrees in electrical and electronic engineering from South China University of Technology, Guangzhou, China in 1984, 1987, and 1990 respectively. He is now a Professor at the Department of Electrical and Computer Engineering of the National University of Singapore.
Professor Li’s research interests include speech information processing, natural language processing, and human-robot interaction. He has published over 300 technical papers. Professor Li has served as Associate Editor (2008-2012), Senior Area Editor (2014-2016), and Editor-in-Chief (2015-2017) of IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH AND LANGUAGE PROCESSING. He has also served as Associate Editor of Computer Speech and Language (2012-), and Springer International Journal of Social Robotics (2008-), and a Member of IEEE Speech and Language Processing Technical Committee (2013-2015). Professor Li is the President of the International Speech Communication Association (ISCA, 2015-2017), the President of Asia Pacific Signal and Information Processing Association (APSIPA, 2015-2016), the President of the Chinese and Oriental Language Information Processing Society (COLIPS, 2011-2013), the Vice President of the Asian Federation of Natural Language Processing (AFNLP, 2015-2016). Professor Li served as the General Chair of ACL 2012 and INTERSPEECH 2014, the Local Arrangement Chair of SIGIR 2008 and ACL-IJCNLP 2009.
Professor Li was the recipient of National Infocomm Awards 2002, President’s Technology Award 2013, and MTI Innovation Activist Gold Award 2015 in Singapore. He was named one of the two Nokia Visiting Professors in 2009 by Nokia Foundation. He is a Fellow of IEEE.

Abstract: We present iCALL, a large scale speech corpus designed to evaluate Mandarin Chinese pronunciation patterns of non-native speakers of European origin. iCALL is a newly developed corpus that consists of 90,841 utterances from 305 speakers with a total duration of 142 hours. The speakers are from diverse linguistic backgrounds (spanning Germanic, Romance, and Slavic first languages). The read utterances are phonetically balanced and are of varying lengths (words, phrases, and sentences). These spoken utterances are phonetically transcribed with lexical tones and perceptually rated with fluency scores by trained native speakers of Mandarin achieving high inter- and intra-rater consistency. In this work, we share our experience in speech corpus design, data collection, and human annotation. Furthermore, we discuss the pronunciation patterns of these non-native speakers by analyzing their phonetic errors, tonal errors, and fluency level distributions. We will also report computational experiments on evaluation of spoken Mandarin Chinese.

◆  Huan Liu, Professor, Arizona State University, USA

Keynote topics: Evaluation Dilemmas in Social Media Research

Short Bio: Dr. Huan Liu is a professor of Computer Science and Engineering at Arizona State University. He obtained his Ph.D. in Computer Science at University of Southern California. He was recognized for excellence in teaching and research in Computer Science and Engineering at Arizona State University. His research interests are in data mining, machine learning, social computing, and artificial intelligence, investigating problems that arise in real-world applications with high-dimensional data of disparate forms. His well-cited publications include books, book chapters, encyclopedia entries as well as conference and journal papers. He serves on journal editorial/advisory boards and numerous conference program committees. He is a Fellow of IEEE and a member of several professional societies.

Abstract: Social media data is steeped with user-generated content and social information. Most of user-generated content can be text and multimedia. Social media is a new source of data and therefore, social media research faces novel challenges. We discuss one of such challenges - evaluation dilemmas. One evaluation dilemma is that there is often no ground truth in evaluating research findings of social media. Without ground truth, how can we perform credible and reproducible evaluation? Another associated dilemma is that we frequently resort to crowdsourcing mechanisms such as Amazon’s Mechanical Turk for evaluation tasks. It costs even if a small group of Turkers is employed. Is it too small? Large-scale evaluation could be very costly. Can we find alternative ways of evaluation that are more objective, reproducible, or scalable? We use case studies to illustrate these dilemmas in NLP related applications and show how to overcome associated challenges in mining big social media data.