(T) I had the unique opportunity to take many years ago the class “Probabilistic Reasoning in Intelligent Systems” from Professor Judea Pearl at UCLA. Professor Pearl pioneered the use of probability theory to deal with uncertainty in artificial intelligence systems. More recently, probability theory and graph theory have been combined to create probabilistic graphical models (PGM) for solving uncertainty, and complexity in many machine learning applications such as medical diagnosis, computer vision, fault analysis, and speech recognition.
This week, I attended a lecture from Dustin Tran, a Ph.D. student at Columbia University and currently a researcher at OpenAI in San Francisco at the Bay Area Probabilistic Programming meet-up at Facebook.
Dustin and many other contributors have developed Edward, a Python library built on TensorFlow for PGM. Edward supports probabilistic modeling, inference, and criticism:
Composable Turing-complete language of random variables
Many data types, tensor vectorization, broadcasting, 3rd party support
Use cases: Graphical models, neural networks, probabilistic programs
- Composable language for hybrids, message passing, data subsampling
- Infrastructure to develop your own algorithms
- Use cases: Black box VI, Hamiltonian MC, stochastic gradient MCMC
- Use cases: Scoring rules, hypothesis tests, predictive checks
Edward is a testbed for fast experimentation and research with probabilistic models, ranging from classical hierarchical models on small data sets to complex deep probabilistic models on large data sets.
Following is Dustin’s presentation on Edward: Edward
If you are interested in contributing to Edward, please see for more details at http://edwardlib.org/contributing
Note 1: A good class on PGM is Stanford Professor Daphne Koller’s on Coursera: PGM’s Representation, Inference, and Learning.
Note 2: The picture above is Facebook building 20 in Menlo Park.
Copyright © 2005-2017 by Serge-Paul Carrasco. All rights reserved.
Contact Us: asvinsider at gmail dot com.
Categories: Machine Learning