BAPP #6: Deep probabilistic models with Edward: The PPL built on TensorFlow
Hosted by Bay Area Probabilistic Programming
Details
IMPORTANT FOR ACCESS: Must RSVP with your email (https://goo.gl/forms/yzRcWdZ8BwXjtBqh1) to gain access. Use shuttle service to get to Facebook.
REGISTER HERE (https://goo.gl/forms/yzRcWdZ8BwXjtBqh1)
Shuttle-service and getting to meeting room: Facebook restricts car access to its campus. It is providing shuttle-service from Palo Alto Caltrain Station: Download pass (https://drive.google.com/file/d/0B3RJ__7YBJgtSUJYMHNkM2FmNnM/view?usp=sharing). The shuttle from Caltrain arrives at building 12. Then take redline shuttle to building 20, zone 4. There you can get a badge from security and get escorted to the meetup room. It takes about 15 minutes from Building 12 to the meetup room.
Location at Facebook: MPK 20, Zone 4, Room Oddjob
Time: Starts 6:00 with drinks, food, and conversation. Presentation starts at 6:30pm, ends at 8PM. (Edit note: changed to 6:30 from 6:30 PM start) Update: the food/drink are served both at 6 pm and 8 pm after the talk. So people can hang out after the talk.
Style: Talk
Presenter: Dustin Tran
Links: Edward (http://edwardlib.org/)
Dustin will talk about his work on Edward, a probabilistic programming language that bridges the gap between probabilistic modeling and traditional deep learning. Edward is a Turing-complete probabilistic programming language that is as flexible and computationally efficient as traditional deep learning. Edward is fast; on a benchmark logistic regression task it is at least 35x faster than Stan and 6x faster than PyMC3. Edward is flexible; the same model can be fit with a variety of composable inference methods, ranging from point estimation to variational inference to MCMC. Edward can reuse the modeling representation as part of inference, facilitating the design of rich variational models and generative adversarial networks.
https://secure.meetupstatic.com/photos/event/d/d/3/d/600_460436637.jpeg
Dustin Tran is a Ph.D. student in Computer Science at Columbia, where I am advised by David Blei and Andrew Gelman. He works in the fields of Bayesian statistics, machine learning, and deep learning. He is most interested in probabilistic models, whether it be in their development, inference, or more generally their foundations for computational and statistical analysis.