(It seems to be a big DARPA-organized effort to create next-generation tools that enable the creation of probabilistic machine learning systems that are significantly more capable and can handle more complex models than current tools allow.)
Re:Probabilistic Programming for Advancing Machine Learning
by jeffreyrogers on Saturday February 07, 2015 @10:42AM
(#9014251)
I've seen a few references to probabilistic programming, but I'm not familiar with it. Can someone who is briefly describe what its advantages are and where it's likely to provide a better solution than more traditional techniques?
Re:Probabilistic Programming for Advancing Machine Learning
by probinso on Saturday February 07, 2015 @06:47AM
(#9013721)
This is a fairly large project, spanning 4 years, so there are quite a bit of moving parts.
There are a group of teams developing new programming languages, and are being funded to do so by darpa. Some of the teams are leveraging their languages, and networking/collaboration opportunities provided them through their participation, to kick off start-ups.
Because success of a language is largely dependent on acceptance by the community, there are also summer schools sessions that are hosted to increase publicity and provide strong feedback to the language development teams on usability.
Those who have interesting and well defined problems in the space of machine learning should consider attending the summer school, or submitting a 'challenge problem' to the Galois team for the language developers to solve.
(It seems to be a big DARPA-organized effort to create next-generation tools that enable the creation of probabilistic machine learning systems that are significantly more capable and can handle more complex models than current tools allow.)
afaik Kinect skeletal pose reconstruction is modeled in PGMs
I thought probabilistic programming has been around for decades.
There are a group of teams developing new programming languages, and are being funded to do so by darpa. Some of the teams are leveraging their languages, and networking/collaboration opportunities provided them through their participation, to kick off start-ups.
Because success of a language is largely dependent on acceptance by the community, there are also summer schools sessions that are hosted to increase publicity and provide strong feedback to the language development teams on usability.
Those who have interesting and well defined problems in the space of machine learning should consider attending the summer school, or submitting a 'challenge problem' to the Galois team for the language developers to solve.