Abstract

Self-supervised learning (SSL) has shown to be vital for advancing research in natural language processing (NLP), computer vision (CV), and speech processing. The paradigm pre-trains a shared model on large volumes of unlabeled data and achieves state-of-the-art for various tasks with minimal adaptation. This talk first introduces the Speech processing Universal PERformance Benchmark (SUPERB), which is a leaderboard to benchmark the performance of SSL model across a wide range of speech processing tasks. The results on SUPERB demonstrates that SSL representations show competitive generalizability across speech processing tasks. This talk will also share some surprising new findings that SSL models pretrained from text are helpful for non-text token sequence classification data, including amino acid, DNA, and music.

Biography

李宏毅 (Hung-yi Lee) received the M.S. and Ph.D. degrees from National Taiwan University (NTU), Taipei, Taiwan, in 2010 and 2012, respectively. From September 2012 to August 2013, he was a postdoctoral fellow in Research Center for Information Technology Innovation, Academia Sinica. From September 2013 to July 2014, he was a visiting scientist at the Spoken Language Systems Group of MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). He is currently an assistant professor of the Department of Electrical Engineering of National Taiwan University, with a joint appointment at the Department of Computer Science & Information Engineering of the university. His research focused on speech technology and machine learning.

Abstract

本演講旨在探討如何運用AI、機器學習、深度學習等智慧演算法與模型建構，定義特定產業或科技領域之知識本體建構、進行專利探勘與布局分析、探索關鍵技術趨勢、甚至智財侵權訴訟判例推薦等 “AI for IP” 的挑戰任務。

-知識本體建構 (eOntology)

-技術功效矩陣 (eTFM)

-智慧判例推薦 (IP infringement recommender)

-Other AI for IP Challenges

Biography

Professor Amy Trappey received her PhD in Industrial Engineering from Purdue University. She started her tenure- tracked faculty career in the Department of Industrial & Manufacturing Systems Engineering, Iowa State University (1990~1992). In 1992, Dr. Trappey returned to Taiwan as an associate professor at National Tsing Hua University (NTHU). She was promoted to a full professor in 1996. From 2003 to 2006, she served as the Department Chairperson of Industrial Engineering and Engineering Management at NTHU and led the department to its initial engineering accreditation approval by IEET. Professor Trappey has also been appointed as the Chair Professor and the Dean of the College of Management at the National Taipei University of Technology (2008/2 ~ 2011/1). She was the Program Director of Industrial Engineering and Management Program at the National Science Council (NSC) from 2007/12 to 2010/12. Professor Trappey's research interests are in the areas of e-business, e-discovery, knowledge engineering, IP management, and engineering asset management. She received awards for her research achievement from organizations, such as NSC/MoST, CIE, and CIIE. Prof. Trappey has also received triple fellowships from the professional societies (ASME Fellow, ISEAM Fellow, and CIIE Fellow), in which she has dedicated her academic career.

Abstract

Speech enhancement (SE) serves as a key component in most speech-related applications. The goal of SE is to enhance the speech signals by reducing distortions caused by additive and convoluted noises in order to achieve improved human-human and human-machine communication efficacy. In this talk, we will review the system architecture and fundamental theories of deep learning-based SE approaches. Next, we will present more recent advances, including end-to-end and goal-driven based SE systems as well as the SE systems with improved architectures and feature extraction procedures. The reinforcement learning and generative adversarial network (GAN)-based SE methods will also be presented. Finally, we will discuss some applications based on the deep learning SE systems, including impaired speech transformation and noise reduction for assistive hearing and speaking devices.

Biography

Yu Tsao (Senior Member, IEEE) received the B.S. and M.S. degrees in electrical engineering from National Taiwan University, Taipei, Taiwan, in 1999 and 2001, respectively, and the Ph.D. degree in electrical and computer engineering from the Georgia Institute of Technology, Atlanta, GA, USA, in 2008. From 2009 to 2011, he was a Researcher with the National Institute of Information and Communications Technology, Tokyo, Japan, where he engaged in research and product development in automatic speech recognition for multilingual speech-to-speech translation. He is currently a Research Fellow (Professor) and Deputy Director with the Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan. He is also a Jointly Appointed Professor with the Department of Electrical Engineering at Chung Yuan Christian University, Taoyuan City, Taiwan. His research interests include assistive oral communication technologies, audio coding, and bio-signal processing. He is currently an Associate Editor for the IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING and IEEE SIGNAL PROCESSING LETTERS. He received the Academia Sinica Career Development Award in 2017, National Innovation Awards in 2018-2021, Future Tech Breakthrough Award 2019, and Outstanding Elite Award, Chung Hwa Rotary Educational Foundation 2019–2020, and 2021 IEEE Signal Processing Society (SPS) Young Author Best Paper Award (corresponding author).

Abstract

Writing a coherent and engaging story is not easy. Creative writers use their knowledge and worldview to put disjointed elements together to form a coherent storyline, and work and rework iteratively toward perfection. Visual storytelling is a task where 5 photos are given to generate a story. A good automated visual storytelling (VIST) model, like people, needs to leverage external knowledge and iteratively generates texts when attempting to create stories. In this talk I will introduce how we achieve this goal by several rounds of model improving, and eventually represent the input image sequence as a story graph linking all the images' elements together using external knowledge. At last, the best storyline is selected to generate the coherent, imaginative and smooth story.

Biography

Lun-Wei Ku is now an associate research fellow in Institute of Information Science, Academia Sinica, adjunct associate professor of national Chiao-Tung university (NCTU), and the secretary-general of Association for Computational Linguistics and Chinese Language Processing (ACLCLP). She received her M.S. and Ph.D. degrees from Department of Computer Science and Information Engineering, National Taiwan University. Her research interests include natural language processing, information retrieval, and computational linguistics. She has been working on sentiment analysis since year 2005 and was the co-organizer of NTCIR MOAT Task (Multilingual Opinion Analysis Task, traditional Chinese side) from year 2006 to 2010. Her international recognition includes Good Design Award Selected (2012), CyberLink Technical Elite Fellowship (2007), IBM Ph.D. Fellowship (2008), and ROCLING Doctorial Dissertation Distinction Award (2009). Other professional international activities she involved include: General Chair, StarSem 2021, Program Chair, StarSem 2019 and ARIS 2019, Best Paper Committee, ACL 2019; Student Workshop Chair, AACL-IJCNLP; Area Chair, ACL 2021, NAACL 2021, ACL 2020, COLING 2020, EMNLP 2019, ACL 2017, CCL 2016, NLPCC 2016, ACL-IJCNLP 2015 and EMNLP 2015; Financial Chair, IJCNLP 2017; Publication Co-Chair, IJCNLP 2013; Publicity Chair, AIRS 2010. She is also active in industrial collaborations and currently working with companies like E-Sun Bank and WinGene.

Abstract

台語是獨特，但瀕臨滅絕的語言。但因為使用者，尤其是年輕一代使用者的數量持續下降。很可能會在幾十年內消失。

為了減緩甚至阻止台語的滅絕，我們從台灣各地，收集大規模台語語音語料庫（目前已語音辨認已收錄600位語者，共300小時，語音合成有四位語者，各十小時），並利用深度學習，開發台語語音辨認跟語音合成技術，希望能日後能有效增進在日常生活中使用台語的機會。

因此，在此次演講中，我將簡介台語語料庫的現況，展示我們目前有的台語音辨認、語言翻譯、語音合成與語音轉換的成果。並希望能跟大家討論國家語言發展法實施後，台語語音AI的機會挑戰。

Biography

Yuan-Fu Liao received the B.S., M.S., and Ph.D. degrees from Department of Communication Engineering, National Chiao Tung University (NCTU), Hsinchu, Taiwan, in 1991, 1993, and 1998, respectively. From January 1999 to June 1999, he was a Postdoctoral Researcher with the Department of Communication Engineering, National Chiao-Tung University. From September 1999 to February 2002, he became a Research Engineer with Philips Research East Asia, Taiwan, Since February 2002, he has been with the Department of Electronic Engineering, National Taipei University of Technology, Taipei, Taiwan, where he is currently a Full Professor. His major research interests are Speech Signal Processing (Speech/Speaker/Language Recognition/Speech Synthesis), Audio Signal Processing (Speech Enhancement, Microphone Array), Natural Language Processing, Machine Learning​ (Deep Learning, Deep Neural Networks).

Abstract

There is a recent boom in utilizing knowledge bases to empower deep learning NLP models. There are two research topics regarding this line: 1) how to automatically and efficiently construct knowledge bases by acquiring a massive amount of information pieces from unstructured texts, and 2) how to represent the acquired or annotated knowledge to inject into deep learning models for different downstream NLP tasks.

In this talk, I will first introduce our recent work on knowledge acquisition, an end-to-end information extraction model based on GCNs to learn named entities and relations from unstructured texts jointly. Following that, I will address the problem of generated noising samples by distantly supervised information extraction and introduce our solution based on a reinforcement learning framework. In the second part of the talk, I will present a new approach to resolve pre-trained LM models’ limited capacity for knowledge pieces.

Biography

Dr. Wei-Yun Ma is an assistant research fellow in the Institute of Information Science at Academia Sinica in Taiwan. He is the principal investigator of the NLP lab at Academia Sinica, a.k.a CKIP lab. His research interests include natural language processing, deep learning, dialogue systems, and knowledge graph construction. Dr. Ma has published more than 35 papers at refereed conferences and journals, including ACL, EMNLP, NAACL, IJCNLP, COLING, WWW, etc. His work received several awards, including the best paper of IALP 2017, the best demo - special mention of WWW 2017, Top1 on the shard task of phrasal arousal prediction held in IJCNLP 2017, Top1 in a Chatbot competition held by PIXNET in 2017, and Top1 in the contest of Chinese word segmentation held by SIGHAN in 2003. His Chinese word segmentation system has being one of the most popular NLP services in Taiwan and has processed several million documents online. Dr. Ma earned his Ph.D. degree from Computer Science Department at Columbia University, New York in 2014. During his studies, he worked on multilingual question answering and machine translation, and also participated in some prestigious projects in the US, such as GALE and STAGES.