human machine interaction

DARPA's plan to speed machine learning development

The Defense Advanced Research Projects Agency wants to make the process of training machine-learning models more efficient.

Currently, training a ML model requires the test data be labeled, which requires humans identify an an image or specific phrases in text that the algorithm should learn to recognize. The more labeled data the system can review, the more complete its training and the better the eventual results.

Amassing enough clean, consistently labeled data to train an algorithm is expensive and time consuming. If, for example, a company wants to analyze user reviews of its products, it will need "at least 90,000 reviews to build a model that performs adequately," according to AltexSoft, a software R&D engineering firm. "Assuming that labeling a single comment may take a worker 30 seconds, he or she will need to spend 750 hours or almost 94 work shifts averaging 8 hours each to complete the task."

DARPA's Learning with Less Labels aims to reduce the amount of labeled data required to build a model by six or more orders of magnitude and cut the number of examples models need to adapt to new environments, according to a recent proposers day announcement.

The plan is to focus on two areas: developing learning algorithms that learn and adapt efficiently and formally characterizing ML problems and proving the limits of learning and adaptation.

DARPA anticipates releasing a broad agency announcement about the program.

More information on the proposers day is available here.

About the Author

Susan Miller is executive editor at GCN.

Over a career spent in tech media, Miller has worked in editorial, print production and online, starting on the copy desk at IDG’s ComputerWorld, moving to print production for Federal Computer Week and later helping launch websites and email newsletter delivery for FCW. After a turn at Virginia’s Center for Innovative Technology, where she worked to promote technology-based economic development, she rejoined what was to become 1105 Media in 2004, eventually managing content and production for all the company's government-focused websites. Miller shifted back to editorial in 2012, when she began working with GCN.

Miller has a BA from West Chester University and an MA in English from the University of Delaware.

Connect with Susan at smiller@gcn.com or @sjaymiller.

inside gcn

  • smart city (jamesteohart/Shutterstock.com)

    Toolkit for building a smart city plan

Reader Comments

Please post your comments here. Comments are moderated, so they may not appear immediately after submitting. We will not post comments that we consider abusive or off-topic.

Please type the letters/numbers you see above

More from 1105 Public Sector Media Group