NLP 280 is a seminar course that features talks from industry experts in the natural language processing (NLP) and artificial intelligence (AI) areas.

The speaker schedule may change without notice, due to changes in speaker availability.

Titles, abstracts, and speaker bios will be made available as the talk date approaches.

Some seminar slots do not have a speaker. Instead, the seminar time will be used for discussion.

The seminar meets weekly on Friday at 2:40 PM, unless otherwise noted.

Date Time Speaker Affiliation Title Abstract Bio
01/08/21 2:40 PM Gokhan Tur Amazon Interactive Learning for Conversational Understanding Recent advances in deep learning based methods for language processing, especially using self-supervised learning methods resulted in new excitement towards building more sophisticated Conversational AI systems. While this is partially true for social chatbots or retrieval based applications, the underlying skeleton of the goal oriented systems has remained unchanged: Still most language understanding models rely on supervised methods with manually annotated datasets even though the resulting performances are significantly better with much less data. In this talk I will cover two directions we are exploring to break from this: The first approach is aiming to incorporate multimodal information for better understanding and semantic grounding. The second part introduces an interactive self-supervision method to gather immediate actionable user feedback converting frictional moments into learning opportunities for interactive learning. Dr. Gokhan Tur is an artificial intelligence researcher, working especially on human/machine conversational language understanding systems. He co-authored about 200 papers published in journals or books and presented at conferences. He is the editor of the book entitled "Spoken Language Understanding" by Wiley in 2011. He received the Ph.D. degree in Computer Science from Bilkent University, Turkey in 2000. Between 1997 and 1999, he was a visiting scholar at the CMU LTI, then the Johns Hopkins University, and the Speech Lab of SRI, CA. At AT&T Research (formerly Bell Labs), NJ (2001-2006) he worked on pioneering conversational systems like "How May I Help You?". He worked for the DARPA GALE and CALO projects at SRI, CA (2006-2010). He was a founding member of the Microsoft Cortana team, and later the Conversational Systems Lab at Microsoft Research (2010-2016). He worked as the Conversational Understanding Architect at Apple Siri team (2014-2015) and as the Deep Conversational Understanding TLM at Google Research (2016-2018). He was a founding area director at Uber AI (2018-2020). He is currently with Amazon Alexa AI.
Dr. Tur is the organizer of the HLT-NAACL 2007 Workshop on Spoken Dialog Technologies, and the HLT-NAACL 2004 and AAAI 2005 Workshops on SLU, and the editor of the Speech Communication Issue on SLU in 2006. Dr. Tur is also the recipient of the IEEE SPS Best Paper Award for 2020, Speech Communication Journal Best Paper awards by ISCA for 2004-2006 and by EURASIP for 2005-2006. He is also the spoken language processing area chair for IEEE ICASSP 2007, 2008, and 2009 conferences and IEEE ASRU 2005 workshop, spoken dialog area chair for HLT-NAACL 2007 conference, NLP applications area chair for HLT-NAACL 2021 conference, and organizer of SLT 2010 workshop.
Dr. Tur is a Fellow of IEEE, and member of ACL and ISCA. He was a member of IEEE Speech and Language Technical Committee (SLTC) (2006-2008), member of the IEEE SPS Industrial Relations Committee (2013-2014) and an associate editor for the IEEE Transactions on Audio, Speech, and Language Processing (2010-2014), and Multimedia Processing (2014-2016) journals. 
01/15/21 2:40 PM Hany Hassan Awadalla Microsoft Pre-trained Multilingual Language Models: from research to production. Transformer-based models are the state-of-the-art for Machine Translation and  Natural Language Understanding (NLU) applications.  Models  are getting bigger and better on various tasks. However, separate models are being developed  to serve various downstream tasks and  Transformer models remain computationally challenging since they are not efficient at inference-time compared to traditional approaches. In this talk we will discuss how to  tackle these limitations to facilitate  deploying such models in  real-life scenarios.  We will discuss how to build cross lingual transformer models  that can be utilized in various  downstream tasks both for generation and classification. We will also discuss  how to  enable efficient runtime  for transformers models  in general to enable  production scenarios. Hany Hassan Awadalla is a Principal Researcher and Research Manager at Microsoft . He is  leading an applied research team to develop and scale cutting edge cross-lingual Natural Language Processing and Machine translation  systems.  He is currently leading the team for developing cross lingual language models platform  that is  being utilized in various products and services across Microsoft. His  work interests are in the areas of Machine Learning and Deep Learning applied to Machine Translation, Natural Language processing, Speech Translation and Semi-Supervised Machine Learning.
01/22/21 1:00 PM Emily Dinan Facebook Safety for Open-Domain Dialogue Agents Over the last several years, neural generative dialogue agents have vastly improved in their ability to carry a chit-chat conversation with humans. However, these models are often trained on large databases from the internet, and as such, may learn undesirable behaviors from this data such as toxic, biased, or otherwise harmful language. In this talk, we will discuss recent advances in conversational AI and present the challenge of creating a safe dialogue agent. We will discuss methods for mitigating toxicity and other safety issues as well as possible avenues for further exploration.

Emily Dinan is a Research Engineer at Facebook AI Research in New York. Her research interests include conversational AI, natural language processing, and fairness and responsibility in these fields. Recently she has focused on methods for preventing conversational agents from reproducing biased, toxic, or otherwise harmful language. Prior to joining FAIR, she received her master's degree in Mathematics from the University of Washington.

 
01/29/21 2:40 PM Sujith Ravi Amazon Large-Scale Neural Graph Learning  Advances in deep learning have enabled us to build intelligent systems capable of perceiving and understanding the real world from text, speech and images. Yet, building real-world, scalable intelligent systems from “scratch” remains a daunting challenge as it requires us to deal with ambiguity, data sparsity and solve complex language & visual, dialog and generation problems. In this talk, I will present powerful neural structured learning frameworks, pre-cursor to widely-popular GNNs, that tackle the above challenges by leveraging the power of deep learning combined with graphs which allow us to model the structure inherent in language and visual data. We use graph-based machine learning as a computing mechanism to design efficient algorithms and address these challenges. Our neural graph learning approach handles massive graphs with billions of vertices and trillions of edges and has been successfully used to power real-world applications at industry scale for response generation, image recognition and multimodal experiences. I will highlight our work on using neural graph learning with a novel class of attention mechanisms over Euclidean and Hyperbolic spaces to model complex patterns in Knowledge Graphs for learning entity relationships, predicting missing facts and performing multi-hop reasoning. Finally, I will describe recent work on leveraging graphs for multi-document news summarization.  Dr. Sujith Ravi is a Director at Amazon Alexa AI where he is leading efforts to build the future of multimodal conversational AI experiences at scale. Prior to that, he was leading and managing multiple ML and NLP teams and efforts in Google AI. He founded and headed Google’s large-scale graph-based semi-supervised learning platform, deep learning platform for structured and unstructured data as well as on-device machine learning efforts for products used by billions of people in Search, Ads, Assistant, Gmail, Photos, Android, Cloud and YouTube. These technologies power conversational AI (e.g., Smart Reply), Web and Image Search; On-Device predictions in Android and Assistant; and ML platforms like Neural Structured Learning in TensorFlow, Learn2Compress as Google Cloud service, TensorFlow Lite for edge devices.
 

Dr. Ravi has authored over 100 scientific publications and patents in top-tier machine learning and natural language processing conferences. His work has been featured in press: Wired, Forbes, Forrester, New York Times, TechCrunch, VentureBeat, Engadget, New Scientist, among others, and also won the SIGDIAL Best Paper Award in 2019 and ACM SIGKDD Best Research Paper Award in 2014. For multiple years, he was a mentor for Google Launchpad startups. Dr. Ravi was the Co-Chair (AI and deep learning) for the 2019 National Academy of Engineering (NAE) Frontiers of Engineering symposium. He was also the Co-Chair for EMNLP 2020, ICML 2019, NAACL 2019, and NeurIPS 2018 ML workshops and regularly serves as Senior/Area Chair and PC of top-tier machine learning and natural language processing conferences like NeurIPS, ICML, ACL, NAACL, AAAI, EMNLP, COLING, KDD, and WSDM.
02/05/21 2:40 PM (No speaker)    Class discussion    
02/12/21 2:40 PM Shereen Oraby Amazon Schema-Guided Natural Language Generation Neural network based approaches to data-to-text natural language generation (NLG) have gained popularity in recent years, with the goal of generating a natural language prompt that accurately realizes some structured data as input. To facilitate the training of these models, researchers have created large datasets of utterances paired with structured inputs, known as meaning representations (MRs), which list the slot and value tokens to be realized. However, the creation of such datasets is an arduous task, and they typically do not include any contextual information that an NLG system can use when trying to generalize across domains. In this talk, we describe the new task of Schema-Guided Natural Language Generation, aimed at addressing the problem of NLG system generalization. We repurpose a dataset from the dialog state tracking community, consisting of a large and rich input schema spanning multiple different attributes, including information about the domain, user intent, and domain-specific slot descriptions. By leveraging this rich schema information as input to NLG models, we show how we are able to produce higher quality outputs in terms of both semantics and diversity, and that we can help models generalize better to new domains. I'm an Applied Scientist at Amazon Alexa AI in Sunnyvale, CA, where I work on natural language generation for Alexa. I completed my PhD at UC Santa Cruz in 2019, working on stylistic control for NLG with Prof. Marilyn Walker at the Natural Language and Dialog Systems Lab. During my PhD, I interned at Amazon, Megagon Labs, and IBM Research, and was a member of UC Santa Cruz’s first Alexa Prize team in 2017. I have worked on various different NLP problems, including sarcasm and figurative language analysis in social media, open domain dialog systems, dialog act classification for customer service, and sentiment analysis for morphologically-rich languages (Arabic). My research interests broadly include dialog systems, conversational AI, and controllable NLG.
02/19/21 2:40 PM François Mairesse Amazon Offline policy optimization for conversational AI This talk will first review the main paradigms in conversational AI, including the modular and end-to-end approaches to statistical dialog systems. We then detail how reinforcement learning offers a way to learn dialog managers by optimizing for task success, rather than merely replicating example dialogs. By trading off exploration and exploitation at run-time, we can collect data that can be used to compare various policies through offline policy optimization, therefore mitigating the need for A/B testing. We report results on the 'Alexa help me find music' voice experience, in which Alexa guides users to find music content.

 Francois Mairesse is a Senior Machine Learning Scientist at Amazon. He joined the Alexa NLU team in 2013 before the launch of the first Echo, and then joined Amazon Music's Conversations team in San Francisco in 2018. Francois did his post-doc in Cambridge's Dialog Systems group, where he focused mostly on statistical modeling for NLU, TTS, and language generation (NLG). He did his PhD in computational linguistics at the University of Sheffield, focusing on statistical methods for detecting human personality cues from spoken language, and generating responses conveying personality along the same set of dimensions.
02/26/21 2:40 PM Antoine Raux Apple Crafted, Learned, and Emergent Conversational Experiences

In this presentation, I will cover different approaches to generate conversational experiences in practical systems today.

Specifically, I will contrast crafted, learned and emergent experiences. What are the underlying technologies of each of them? What are their strengths and weaknesses?Why is it so difficult to combine them? I hope to raise awareness and maybe interest in academic research that could help bridge these gaps and provide the most natural and effective experiences for users of conversational AI, highlighting some recent research that might pave the way to human-like interactions.

 Antoine Raux is an engineering manager at Apple Siri. Previously, Antoine was co-founder of a conversational AI startup (b4.ai), worked at Facebook Applied Machine Learning, and Honda Research Institute. All his career has been focused on building practical conversational AI systems. He holds a master degree from Kyoto University and a PhD in Language Technologies from Carnegie Mellon University.
03/05/21 2:40 PM Jeff Adams Cobalt Speech Future Trends in Commercial Speech Processing Voice technology has dramatically improved in recent years and is rapidly being adopted by consumers. At Cobalt Speech, we get a bird’s eye view of the commercial landscape, see shifts in underlying technology from the research community, and hear first-hand what organizations are building. In this session we’ll tell the story behind Cobalt and shine a light on some of the broader industry trends that we expect to continue into the next decade. Jeff Adams is a leading figure in the world of commercial speech technology.  He has been managing top-level speech & NLP technology research for more than 20 years, at Kurzweil AI, Nuance / Dragon, Yap, and Amazon, before founding Cobalt Speech & Language.  He led the teams that developed the core technology for Dragon NaturallySpeaking, Yap Voicemail, and Amazon Alexa. At Cobalt, Jeff has assembled a team of elite speech scientists & engineers to build custom applications to meet the needs of their clients.  Jeff is the author of 25 patents and several published research papers and serves on the boards of several companies.
03/12/21 2:40 PM (No speaker)   Class discussion