Deep Learning (DL)

archived

DARPA is soliciting innovative research proposals in the area of deeply layered machine learning or, simply, Deep Learning (DL). Over the course of the envisioned program, performer teams will build a universal machine learning engine that uses a single set of methods in multiple layers (at least three internally) to generate progressively more sophisticated representations of patterns, invariants, and correlations from data inputs. The engine is expected to be applicable to multiple input modalities given only changes to the inputs’ preprocessing, and is expected to be able to learn important characteristics of the inputs and produce useful representations solely on the basis of unlabeled inputs. Accomplishing many of the tasks set by the program will require that the engine be able to produce and utilize sophisticated spatio-temporal representations.

High