ADP: Autonomous Deep Perception

Autonomous systems acquire massive amounts of sensor and communications data over the course of their potentially lengthy missions. Ideally, such systems would incorporate current and historical data into their decision making processes to generalize from experience and avoid repetitive errors. However, the sheer quantity of data gathered can make storage and processing of an unfiltered data stream practically difficult. As a result, many current autonomous systems utilize only recent sensor data. In contrast, biological systems quickly summarize highly complex sensory information streams into a lifetime of well-organized memories, which can be quickly accessed to affect current reasoning tasks. Lynntech"s Autonomous Deep Perception (ADP) system will use deep belief neural networks, coupled with life-long learning methods, with the goal of allowing autonomous systems to quickly generate and archive small, salient, and highly-accessible representations of sensor information. Using this transformed, highly relevant view of the incoming data, autonomous systems can identify and focus on mission-relevant input, as well as quickly scan through historical, compactly-represented data archives to apply past experience to current decisions in real time.