초록 ▼

A Forward Feed Neural Network is disclosed using data flow techniques on a data flow microprocessor. As a result of this invention, a neural network is provided that has the capacity of “learning”to distinguish among patterns of data which may differ recognizably from idealized cases, and is able to

A Forward Feed Neural Network is disclosed using data flow techniques on a data flow microprocessor. As a result of this invention, a neural network is provided that has the capacity of “learning”to distinguish among patterns of data which may differ recognizably from idealized cases, and is able to perform pattern recognition faster while utilizing less memory and fewer clock cycles than neural networks implemented on sequential processors. This implementation is simpler and faster because of an inherent similarity between the flow of information in the brain and in data flow architecture.

대표청구항 ▼

A device for performing pattern recognition comprising: a data flow processor; input means for inputting a set of data tokens into said data flow processor, each data token containing a data value; copy means, within said data flow processor, linked to said input means receiving said set of data tok

A device for performing pattern recognition comprising: a data flow processor; input means for inputting a set of data tokens into said data flow processor, each data token containing a data value; copy means, within said data flow processor, linked to said input means receiving said set of data tokens for making multiple copies of said set of data tokens; a plurality of weighting means, within said data flow processor, each linked to said copy means and receiving a copy of said set of data tokens for associating a predetermined weight value to each data value of said set of data tokens and for generating a weighted product of each of said data values with said predetermined weight value associated with each of said data values; and a plurality of neuron means, within said data flow processor, each linked to one of said weighting means receiving therefrom said weighted products provided, for generating a sum of said weighted products by summing said weighted products, for determining within said data flow processor a first output value whenever said sum is greater than or equal to a predetermined threshold value, and for determining within said data flow processor a second output value whenever said sum is less than said predetermined threshold value, wherein each of said plurality of neuron means comprises means for receiving said weighted products; summing means for receiving two values and for generating a partial sum, which is a sum of said two values, by summing said two values; queue means receiving said partial sum and said weighted products for holding said partial sum and said weighted products received and for passing a first value and a second value to a summing means, said first value and said second value corresponding to the earliest two values held by the queue means; convolving means for outputting a predetermined queue means element corresponding to a full neural sum; comparing means receiving said full neural sum for reading said predetermined threshold value from a storage location and for generating a predetermined third output value based on the arithmetic difference between said full neural sum and said predetermined threshold value.