MTL Machine Learning: Hiver 2017 Winter Meetup

1. Challenges of building AI-powered chatbots that yield business value

Erdem Özcan, Head of Research @ Automat

Chatbots are at the intersection of messaging and artificial intelligence. Combining these two trends to build solutions that yield real business value is way more challenging than simply plugging together messaging platforms with natural language understanding APIs. In this talk, Erdem will share some of these challenges and the lessons they have learned during Automat's journey.

2. Highlights from the NIPS conference

Nicolas Chapados, Chief Science Officer @ Element AI

NIPS (Neural Information Processing Systems) is the flagship conference in the areas of learning algorithms and neural computation. Nicolas will present some of the most interesting new research results and insights that surfaced during the most recent edition of this conference.

3. Lightning Community Announcements

Do you have a short announcement of relevance to the local Machine Learning community to make? Send an email to seb -at - elementai.com.

4. Conversations: Recent Breakthroughs

2016 was an exciting year for Machine Learning. Let’s gather in smaller groups, unconference-style, and together talk about some of the exciting results that have recently come to the fore. Discussion leaders will each introduce a topic and facilitate a conversation around the techniques and the promising directions they opened.

• Libratus (https://arxiv.org/abs/1701.01724) made history by defeating four of the world’s best professional poker players in a 20-day poker competition at Rivers Casino in Pittsburgh.

• AlphaGo (https://www.theatlantic.com/technology/archive/2016/03/the-invisible-opponent/475611/) beat Lee Sedol, a Go world champion, 4-1 last March.

• Wavenets (https://www.youtube.com/watch?v=CqFIVCD1WWo) generate speech which mimics human voice and sounds more natural than the best existing Text-to-Speech systems.

• Image Super-Resolution through Deep Learning (https://github.com/david-gpu/srez) turns images into higher-resolution versions of themselves with plausible sharp features.

• The phased LSTM network (https://arxiv.org/abs/1610.09513) adds a time gate to achieve faster convergence than a regular long short-term memory (LSTM) network on tasks which require learning of long sequences.

• LipNet (https://www.technologyreview.com/s/602949/ai-has-beaten-humans-at-lip-reading/) was able to identify 93% percent of words correctly from clips of people reading three-second sentences. Human lip-reading volunteers asked to perform the same tasks identified just 52%.

• Artistic Style Transfer for Videos (https://www.youtube.com/watch?v=Uxax5EKg0zA) builds on style transfer for images: given an input video and a reference image, it creates a video that imitates the style of the reference image with the content of the input video.

• Deep Q-learning (https://www.youtube.com/watch?v=V1eYniJ0Rnk) plays Atari games and improves itself to a superhuman level.

• StackGAN (https://www.youtube.com/watch?v=rAbhypxs1qQ) synthesizes high resolution photo-realistic images from text descriptions.

Speaker Bios

Erdem Özcan is Co-founder and Head of Research of Automat, a seed stage conversational AI startup that is focused on enabling businesses to have personalized, one-on-one conversations with their customers. The cofounders and team collectively have 17 patents in AI and NLP related fields. Automat has received investments from Relay and Real Ventures, the Slackbot fund, USAA and You & Mr. Jones. Their advisory board consists of technology trend spotter Tim O’Reilly, former Chief Creative Officer of TellMe and Nuance Gary Clayton, Richard Socher Chief Scientist at Salesforce, and conversational commerce pioneer Chris Messina. Erdem holds a PhD in Computer Science from Grenoble Institute of Technology and has worked as a researcher in larger companies for more than 10 years.

Nicolas Chapados is Chief Science Officer at Element AI, the world’s leading applied research lab in AI. In 2001 he co-founded ApSTAT Technologies, a machine learning technology transfer firm, with his thesis advisor Yoshua Bengio. He also co-founded two spin-off companies: Imagia, to provide AI-based actionable predictive oncology analytics atop medical imaging data, and Chapados Couture Capital, a Quebec-registered quantitative asset manager. He holds a PhD in Computer Science from University of Montreal.