Neural network that does not require repetitive training

• US 5,517,667 A
• Filed: 06/14/1993
• Issued: 05/14/1996
• Est. Priority Date: 06/14/1993
• Status: Expired due to Fees

• Current Assignee: Freescale Semiconductor, Inc. (NXP Semiconductors NV)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Wang, Shay-Ping T.
• Primary Examiner(s): Davis, George B.
• Application Number:
  US08/076,601
• Time in Patent Office: 1,065 Days
• Field of Search: 395/21, 395/23, 395/24, 395/22, 395/27
• US Class Current: 706/31
• CPC Class Codes: G06F 2207/5523 Calculates a power, e.g. the square, of a number or a function, e.g. polynomials
• CPC Class Codes: G06F 7/552 Powers or roots , e.g. Pythagorean sums
• CPC Class Codes: G06N 3/063 using electronic means

Abstract

A neural network, which may be implemented either in hardware or software, is constructed of neurons or neuron circuits each having only one significant processing element in the form of a multiplier. A hidden neuron in the neural network generates an output based on the product of a plurality of functions. The neural network utilizes a training algorithm which does not require repetitive training and which yields a global minimum to each given set of input vectors.

Claims

1. A neural network having a plurality of network inputs and at least one network output, said neural network comprising:

a plurality of input nodes, each input node coupled to a corresponding one of said plurality of network inputs and generating corresponding outputs;

a plurality of hidden neurons, each hidden neuron having a plurality of neuron inputs and generating an output based on the product of a plurality of functions, wherein each of said plurality of functions is proportional to one of said plurality of neuron inputs raised to the power of one of a plurality of constants;

means for distributing said output of each input node to one of said neuron inputs of at least one of said plurality of hidden neurons; and

means for summing outputs of said hidden neurons and generating said at least one network output.

2. The neural network recited in claim 1 wherein each of said plurality of constants is an integer.

3. The neural network recited in claim 2 wherein said at least one network output is based on a weighted sum of the outputs of said hidden neurons.

4. A neural network having a plurality of network inputs and a plurality of network outputs, said neural network comprising:

an input node for receiving successive ones of said plurality of network inputs and generating corresponding node outputs;

a plurality of hidden neurons, each hidden neuron having at least one neuron input and generating a neuron output based on a product of a plurality of functions, each of said plurality of functions being a function of said at least one neuron input;

means for distributing said node outputs to said at least one neuron input of each of said plurality of hidden neurons; and

means for summing neuron outputs and generating said plurality of network outputs, said means for summing including a plurality of summers each summing the outputs of predetermined ones of said hidden neurons and generating said network outputs.

5. The neural network recited in claim 4 wherein said hidden neurons comprise a linear transfer function.

6. The neural network recited in claim 4 wherein the outputs of at least some of said hidden neurons may be summed by more than one of said summers.

7. The neural network recited in claim 4 wherein said neural network is contained on at least one integrated circuit.

8. The neural network recited in claim 4 wherein said neural network is contained in a computer program.

9. The neural network recited in claim 4 wherein each of said plurality of functions is proportional to said at least one neuron input raised to the power of at least one constant.