Universal Computation With Low-Complexity Wireless Relay Networks
The authors propose a method for enabling complex computations in a network of low-complexity wireless devices. By utilizing multi-hop relaying, such devices can form the wireless equivalent of an Artificial Neural Network (ANN). They provide a method for programming the network functionality in a decentralized fashion and demonstrate the robustness of wireless ANNs against node failures and imperfections. Applications of this scheme exist in low-complexity sensor networks, where elaborate calculations can be carried out in a distributed fashion, or for creating powerful ANNs with very high degrees of interconnectivity realized by the wireless medium.