Science in Action: DeepLearning@StitchFix, InfoViz@Berkeley,FeedRanking@Quora
Details
Dear Friends,
OpenTable is hosting its inaugural OpenScience Meetup! Come and learn about Deep Learning tricks from Stitch Fix, InfoViz from UC Berkeley, and applied machine learning in personalized feed optimization from Quora.
6:00-6:30 Meet and Greet. Refreshments.
6:30 - 8:00 Talks
8:00 - 8:15 Discussions
------------------------TALKS -------------------
6:30 - 7:00 Deep Learning's Bag of Tricks
Speaker: Christopher E Moody, Stitch Fix
Abstract:
At first glance, deep learning can seem like a daunting endeavor. I'll cover a grab bag of practical tricks & techniques to get you started if you're new or bring you up the state-of-the-art if you're experienced. Time permitting, I'll cover the basics in the framework Chainer (http://chainer.org/) (an alternative to Theano or Torch7), distributed representations, the reparameterization trick (that connects deep learning to disciplined statistics), and LSTMs (that learn sequences). At Stitch Fix we use this set of techniques to better understand client comments and respond with an experience curated for them personally, and I think there are exciting parallels at OpenTable!
Speaker Bio:
After a degree in Physics from Caltech and Ph.D in astrophysics & supercomputing from UCSC, Chris is now at Data Labs at Stitch Fix doing machine learning & statistics. In addition to deep learning, he is currently knee-deep in Gaussian Processes, t-SNE, word2vec, tensor decompositions, and factorization machines -- so if any of this sounds interesting to you let's grab some coffee! (He can be reached at moody@stitchfix.com )
--------------------------------------------------------------------
7:00 - 7:30 Infoviz, The Next Generation
Marti Hearst, Professor, UC Berkeley
Information visualization has escaped the research lab and is now heavily used by practitioners across a wide spectrum of fields. New software tools and programming frameworks appear on a monthly basis. New design paradigms are rapidly gaining acceptance and evolving. As the role of information visualization grows and changes in the world of practice, new methods are needed to teach this dynamic topic.
Inspired by Eric Mazur (https://en.wikipedia.org/wiki/Eric_Mazur)'s writings about how he transformed instruction in physics courses with active and peer learning, and by the technology for active learning used in MOOCs, for the past two years I have created my own blend of these ideas and introduced them into my information visualization course. In this talk, I will describe this new style of teaching and give examples of the outcomes, including a menu data visualization project very relevant to OpenTable. With luck some of the Masters students who did the projects will join me in the presentation.
Speaker bio:
Prof. Marti Hearst (https://en.wikipedia.org/wiki/Marti_Hearst) has taught Information Visualization since 1998 at UC Berkeley. She learned Infoviz while a researcher at Xerox PARC in the early 1990's and her research focus in this field is on text visualization. She wrote the book "Search User Interfaces" in 2009 (Cambridge University Press) and received student-initiated Excellence in Teaching Awards in 1999, 2002, 2014 and 2015.
--------------------------------------------------------------------
7:30 - 8:00 Machine Learning for Quora’s Feed
Alberto Bietti, Quora
Abstract
At Quora our mission is to “share and grow the world's knowledge”. We want to do this by getting the right questions to the right people to answer them, but also by getting the existing answers to people who are interested in them. An essential component for making this possible is Quora’s home feed, which allows us to distribute questions and answers to users in a personalized and engaging way. Machine learning is crucial for making the feed experience more engaging. I will talk about how our feed machine learning system works, and will give an overview of other Quora products which leverage machine learning.
Speaker Bio:
Alberto Bietti is a software engineer at Quora, where he works on machine learning, systems and infrastructure for Quora's feed and other related ranking products. He holds an MS in applied math, machine learning and vision from Ecole Normale Supérieure, as well as an MS in engineering and applied math from Mines ParisTech. Before Quora, he worked on ads optimization at Facebook, and did research in machine learning and its applications to computer vision and audio signal processing at Caltech, INRIA and IRCAM.
--------------------------------------------------------------------
See you there,
The OpenScience Team!