Principal Investigator Vikash Mansinghka
Co-investigators Tejas Dattatraya Kulkarni , Alexey Radul
Project Website http://probcomp.csail.mit.edu.ezproxy.canberra.edu.au/
The MIT Probabilistic Computing Project is hosted by MIT's Computer Science and Artificial Intelligence Laboratory and Department of Brain and Cognitive Sciences. Work is generously supported by research contracts with Defense Advanced Research Projects Agency (DARPA) -- under the XDATA and PPAML programs -- the Office of Naval Research (ONR) and the Army Research Laboratory (ARL), and Shell Oil, as well as gifts from Analog Devices and Google. The views expressed on this website and in our research are our own, and do not necessarily reflect the views of our government or corporate sponsors.
We build probabilistic computing systems that exploit uncertain knowledge to learn from data, infer its probable causes, make calibrated predictions and choose effective actions. We also study the computational principles and building blocks needed to design, implement and analyze these systems, drawing on and contributing to an emerging integration of key ideas from probability theory and computer science. Our research includes work on machine learning and artificial intelligence fundamentals, as well as applications to modeling human cognition and to intelligent data analysis.
So far, this work has yielded new general-purpose probabilistic programming technology and intentionally stochastic (but still digital) hardware for real-time Bayesian inference. It has also yielded academic and commercial Bayesian database systems that automate the analysis of high-dimensional data tables.
Research
(*) Venture --- an interactive, Turing-complete probabilistic programming platform descended from Church(*) BayesDB --- a Bayesian database that lets users query the probable implications of their data, and solve basic data science problems without training in statistics(*) CrossCat --- a nonparametric Bayesian machine learning method for analyzing high-dimensional data tables(*) Generative probabilistic graphics programming --- using probabilistic programming and computer graphics to solve 2D and 3D vision problems