No abstract

The focus of this paper is on multitask learning over adaptive networks where different clusters of nodes have different objectives. We propose an adaptive regularized diffusion strategy using Gaussian kernel regularization to enable the agents to learn about the objectives of their neighbors and to ignore misleading information. In this way, the nodes will be able to meet their objectives more accurately and improve the performance of the network. Simulation results are provided to illustrate the performance of the proposed adaptive regularization procedure in comparison with other implementations.

Abstract:In this study, we propose a novel method to investigate P300 variability over different trials. The method incorporates spatial correlation between EEG channels to form a cooperative coupled particle filtering method that tracks the P300 subcomponents, P3a and P3b, over trials. Using state space systems, the amplitude, latency, and width of each subcomponent are modeled as the main underlying parameters. With four electrodes, two coupled Rao-Blackwellised particle filter pairs are used to recursively estimate the system state over trials. A number of physiological constraints are also imposed to avoid generating invalid particles in the estimation process. Motivated by the bilateral symmetry of ERPs over the brain, the channels further share their estimates with their neighbors and combine the received information to obtain a more accurate and robust solution. The proposed algorithm is capable of estimating the P300 subcomponents in single trials and outperforms its non-cooperative counterpart.

The emergence of distributed and complex networks has altered the field of information and data processing in the past few years. In distributed networks, the connected neighboring nodes can cooperate and share information with each other in order to solve particular tasks. However, in many applications the agents might be reluctant to share their true data with all their neighbors due to privacy and security constraints. In this paper, we study the performance of multitask distributed networks where sharing genuine information is subject to a cost. We formulate an information credibility model which results in the probability of sharing genuine information at each time instant according to the cost. Each agent then shares its true information with only a subset of its neighbors while sending fabricated data to the rest according to this probability. This behavior can affect the performance of the whole network in an adverse manner especially in cases where the cost is high. To overcome this problem, we propose an adaptive reputation protocol which enables the agents to evaluate the behavior of their neighbors over time and select the most reputable subset of neighbors to share genuine information with. We provide an extensive simulation-based analysis to compare the performance of the proposed method with several other distributed learning strategies. The results show that the proposed method outperforms the other learning strategies and enables the network to have a superior performance especially when the cost of sharing genuine information is high.

No abstract