Dr. JunLing Hu, Director of Applied AI, LivePerson, Inc
In the last 2 years, we are seeing rapid development of conversational AI technology. Current chatbots can speak coherently and meaningfully, almost matching human performance. Are we close to cracking Turing Test? What are the other challenges we face today? In this talk, I will review the state-of-the-art dialog systems, and their underlying deep learning architecture. I will review two types of dialog systems: Open-domain chatbots that can make conversation on any subject, and task-oriented chatbots that serve the user for specific needs (such as ordering food or booking a ticket). I will discuss the success and challenges in both domains. Finally, I will share our work at LivePerson for providing conversational AI solutions to more than 18,000 enterprise customers.
Speaker Bio:
Dr. Junling Hu is the Director of Applied AI at LivePerson, Inc, where she leads a team on natural language understanding and conversational AI. She is the author of the book "The Evolution of Artificial Intelligence". and a recipient of the National Science Foundation CAREER award. Dr. Hu was the director of data mining at Samsung Electronics from 2013 to 2015, where she led a team to create intelligent AI products for large-scale recommender systems. Before Samsung, Dr. Hu has led data science team at PayPal, eBay, and Bosch, to deliver innovative NLP products and machine learning solutions. Dr. Hu received her PhD in AI from University of Michigan in Ann Arbor.
Introducing the LivePerson Conversational Cloud
We are hiring! • Data scientist of all levels
For more details:
0:00 SFBayACM Chapter Intro
3:39 Speaker Intro
4:56 Future of Conversational AI
8:34. Building a Chatbot: Traditional Architecture
11:51 New architecture: End-to-end systems Transformer
16:57 T5 pre-training • Train on Colossal Clean Crawled Corpus (C4)
17:50 BART (Bidirectional and Auto-Regressive Transformer)
20:41 Two types of Dialogue Systems Open-domain dialog 2. Task Oriented Dialog (TOD)
29:26 BlenderBot 2.0
39:06. End-to-end Task-Oriented Dialog (TOD) system
43:20 The Ticket Talk Methods
1:01:25 LivePerson introduction
1:04:19 Q & A
