Under the Radiofrequency (RF) Machine Learning Systems (RFMLS) program, DARPA developed foundational technologies that enable autonomous RF systems to learn directly from data. By applying deep neural networks (DNNs) directly to digitized RF signals, RFMLS extracted more discriminating signal features and accelerated the implementation of detectors relative to traditional signal processing approaches.

RFMLS demonstrated the value of attentional algorithms that focus on important signals. Extending this concept of attention to emerging runtime reconfigurable processors may enable edge systems to process higher signal densities across wider bandwidths.

RFMLS focused on supervised learning across labelled datasets, but these approaches have reduced applicability when systems must operate in highly uncertain environments. Sample efficient learning algorithms coupled with automated design of experiences has the potential to improve the robustness of autonomous RF systems that must operate in environments that are difficult to specify at design time.

This talk will further discuss key results of the RFMLS program, open related technical challenges, and corresponding potential areas for future research.