DARPA wants unique automated tools to rapidly make computers smarter

DARPA program looks to bolster Artificial Intelligence tools to make machine learning easier

darpa
Researchers at DARPA want to take the science of machine learning -- teaching computers to automatically understand data, manage results and surmise insights -- up a couple notches.

Machine learning, DARPA says, is already a the heart of many cutting edge technologies today, like email spam filters, smartphone personal assistants and self-driving cars. "Unfortunately, even as the demand for these capabilities is accelerating, every new application requires a Herculean effort.  Even a team of specially-trained machine learning experts makes only painfully slow progress due to the lack of tools to build these systems," DARPA says.

[NEWS: The year in madly cool robots]

[MORE: DARPA system to blend AI, machine learning to understand mountain of text]

While machine learning has come a long way, the agency want to develop way more advanced tools under a program it will detail next month called Probabilistic Programming for Advanced Machine Learning (PPAML).  To that end the program has some lofty goals including:

  • dramatically increasing the number of people who can successfully build machine learning applications
  • making machine learning experts radically more effective
  • enabling new applications that are impossible to conceive of using today's technology.

 In support of this overarching goal, the program wants to make machine learning model code shorter to reduce development time as well as reduce the level of expertise needed to build machine learning applications.

"We want to do for machine learning what the advent of high-level program languages 50 years ago did for the software development community as a whole," said Kathleen Fisher, DARPA program manager.  "Our goal is that future machine learning projects won't require people to know everything about both the domain of interest and machine learning to build useful machine learning applications. Through new probabilistic programming languages specifically tailored to probabilistic inference, we hope to decisively reduce the current barriers to machine learning and foster a boom in innovation, productivity and effectiveness."

If these goals aren't challenging enough, the agency went onto detail some other issues.  "Anticipated research challenges in this area include:

  • advancing the theory of probabilistic programming;
  • discovering new inference algorithms that are more efficient, more accurate, more predictable, or more generalizable;
  • discovering novel representations that support more efficient, more accurate, more predictable, or more generalizable inference;
  • developing inference algorithms that work over streaming data or have better scaling properties; and
  • developing techniques for assessing model fitness for a particular data set.

Any systems developed under PPAML will be evaluated using a collection of challenge problems that span the range of machine learning applications, DARPA said. Each system will be evaluated on how well it performs on all challenge problems. In addition, to evaluate the effectiveness of program technologies in enabling the rapid creation of new machine learning applications by domain experts, the program will include annual "Summer Schools" of two to four weeks.

DARPA will host a PPAML Proposers' Day on Wednesday, April 10, 2013. For details, look here.

 Follow Michael Cooney on Twitter: nwwlayer8 and on Facebook

Check out these other hot stories:

25 crazy and scary things the TSA has found on travelers

Laser pointers produce too much energy, pose risks for the careless

US intelligence agency wants to know how to use alternate reality gaming to bolster research

NASA's Inspector General paints bleak picture for agency projects, IT security

NASA: Mars rock sample shows Red Planet could have supported life

NASA exploring groundbreaking space network to sustain large data dumps and trips to the moon, Mars

Robots get an open source Web-based helpline

Editors' Picks
Join the discussion
Be the first to comment on this article. Our Commenting Policies