Federated Learning based Cooperative Sensing for Efficient Threat Detection (FedSense)

Knowledge Based Systems, Inc. (KBSI) proposes to extend the federated learning (FL) architecture to support flexible and integrated interceptor discrimination and decision making. Specifically, KBSI will research, design, and demonstrate the Federated Learning based Cooperative Sensing for Efficient Threat Detection (FedSense) framework that allows models that are learnt locally at terminal nodes to be efficiently shared, transformed and accepted as whole or in part based on the quality and value offered by each model. The accepted heterogenous models are then fused together into an integrated model using federated transfer learning method. The FedSense methods helps to learn from terminal nodes (sensors) with large volumes of data and helps to boost the performance of terminal node which has limited amount of training data. Approved for Public Release | 21-MDA-11013 (19 Nov 21)