First Demonstration of Opto-Electronic Reservoir Computing
Technology Review (12/02/11)
Universite Libre de Bruxelles researchers have developed a form of computing that exploits feedback loops to perform extremely fast analog calculations. The researchers found that a nonlinear feedback mechanism is basically an information processor, because it takes a certain input and processes it to generate an output. The feedback loop is a kind of memory that stores information about the system's recent history, making this form of processing an analysis of just a small segment of the recent past. The researchers, led by Yvan Paquot, are working on reservoir computing, which consists of a large number of nodes that are randomly connected. Each node is a kind of non-linear feedback loop, and the inputs are fed into the random nodes in the reservoir, while the outputs are taken from other randomly chosen nodes. The researchers say the reservoir network is similar to a neural network, except the output signals are weighted during training, which makes the process much simpler than with a neural network. "Our experiment is the first implementation of reservoir computing fast enough for real time information processing," Paquot says.
Technology Review (12/02/11)
Universite Libre de Bruxelles researchers have developed a form of computing that exploits feedback loops to perform extremely fast analog calculations. The researchers found that a nonlinear feedback mechanism is basically an information processor, because it takes a certain input and processes it to generate an output. The feedback loop is a kind of memory that stores information about the system's recent history, making this form of processing an analysis of just a small segment of the recent past. The researchers, led by Yvan Paquot, are working on reservoir computing, which consists of a large number of nodes that are randomly connected. Each node is a kind of non-linear feedback loop, and the inputs are fed into the random nodes in the reservoir, while the outputs are taken from other randomly chosen nodes. The researchers say the reservoir network is similar to a neural network, except the output signals are weighted during training, which makes the process much simpler than with a neural network. "Our experiment is the first implementation of reservoir computing fast enough for real time information processing," Paquot says.
No comments:
Post a Comment