part of the Probabilistic Programming for Advanced Machine Learning
(PPAML) DARPA program. PPAML's goal is to further the research towards
frameworks in which the user can specify probabilistic generative models
using a Turing-complete programming language, and query this model in
The PRAiSE project provides research in lifted inference for PPAML.
Lifted inference is a symbolic way of performing inference directly at
the higher-order level in which probabilistic programs are specified, as
opposed to grounding them to propositional graphical models. The latter
option is better understood today, but much slower. PRAiSE seeks to make
probabilistic programs inference scalable.
The main advantages of PRAiSE will be the solving of expressive probabilistic models, specified in well-known, convenient programming languages, producing comprehensible output, explanations, and debugging information.
Typical areas of applications will be natural language processing, vision, planning, and common-sense reasoning.