Cooperative Classification over Distributed Networks

Distributed networks observe events from various perspectives, enabling classification tasks using complementary data received across nodes. However, a centralized strategy that transmits raw data or heavy local features to a central hub can lead to scalability and communication burdens. In this work, we present a cooperative classification scheme designed for distributed networks. In this scheme, each node learns a classifier using its locally extracted features from data and compressed features received from other nodes. Furthermore, the compression process itself is trainable and adapted based on the local data at each node. To validate the efficacy of our method, we conducted simulations focused on acoustic target recognition within underwater sensor networks. Our findings demonstrate the practical advantages of our cooperative classification scheme, particularly in scenarios where efficient information sharing among network nodes is crucial.