부산대학교 도서관

In a distributed algorithm several processing nodes communicate with each other and incorporate in order to reach a common goal. Each unit has its own locally-observed environment while it can either help or mislead other units when sharing its information. This could be either intentional or simply due to the observation noise. Previous works mainly focus on locating the agents properly and assigning weights based on stationary environment in order to minimize the impact of noisy nodes. We, however, propose a method capable of assigning combination weights in accordance with momentary performance of the neighboring nodes. In order to demonstrate this capability, the proposed algorithm has been applied to well-known diffusion LMS algorithms. Unlike uniform weighting policy, the proposed method is robust to changes caused by malfunctioning nodes.