Method and apparatus for processing data in a neural network

• US 6,041,322 A
• Filed: 04/18/1997
• Issued: 03/21/2000
• Est. Priority Date: 04/18/1997
• Status: Expired due to Term

• Current Assignee: Transpacific IP Group Limited
• Original Assignee: Industrial Technology Research Institute
• Inventors: Meng, Wan-Yu, Hsu, Fang-Ru, Chang, Cheng-Kai, Lee, Ming-Rong, Chang, Hwai-Tsu
• Primary Examiner(s): Hafiz, Tariq R.
• Assistant Examiner(s): STARKS, WILBERT L
• Application Number:
  US08/839,818
• Time in Patent Office: 1,068 Days
• Field of Search: 706/15, 706/16, 706/25, 706/26, 706/31, 706/38, 706/41, 706/43
• US Class Current: 706/43
• CPC Class Codes: G06N 3/048 Activation functions
• CPC Class Codes: G06N 3/063 using electronic means

Abstract

A digital artificial neural network (ANN) reduces memory requirements by storing sample transfer function representing output values for multiple nodes. Each nodes receives an input value representing the information to be processed by the network. Additionally, the node determines threshold values indicative of boundaries for application of the sample transfer function for the node. From the input value received, the node generates an intermediate value. Based on the threshold values and the intermediate value, the node determines an output value in accordance with the sample transfer function.

Claims

1. An apparatus for processing data in a neural network having a stored sample transfer function, comprising:

a receiver configured to receive at least one input value representing information to be processed by the network;

a threshold processor configured to determine threshold values indicating boundaries for application of the sample transfer function for a node, the threshold processor comprising;

a mapper configured to receive a value representing a specific transfer function for the node and map the received value to identify threshold values for the specific transfer function;

a generator configured to generate an intermediate value from the input value for the node; and

an output processor configured to determine an output value for the node based on the threshold values and the intermediate value in accordance with the sample transfer function.

2. The apparatus of claim 1 wherein the output processor comprises:

a comparator configured to compare the intermediate value to the threshold values and to determine the output value for the node in accordance with the sample transfer function if the intermediate value is between the threshold values.

3. The apparatus of claim 2 wherein the threshold values include upper and lower threshold values;

wherein the comparator comprises;

a first subcomparator configured to set the output value for the node equal to a first saturation value if the intermediate value is less than the lower threshold value; and

a second subcomparator configured to set the output value for the node equal to a second saturation value if the intermediate value is greater than the upper threshold value.

4. The apparatus according to claim 1 wherein the output processor comprises:

a shifter configured to shift the intermediate value.

5. A computer program product comprising:

a computer usable medium having computer readable code embodied therein for processing data in a neural network, the computer usable medium comprising;

a receiving module configured to receive an input value representing information to be processed by the network;

a threshold processing module configured to determine threshold values indicating boundaries for application of a sample transfer function for a node, said threshold processing module comprising;

a defining module configured to define a value representing a specific transfer function for the node; and

a mapping module configured to map the defined value to identify threshold values for the specific transfer function;

a determining module configured to determine an intermediate value from the input value for the node; and

an output processing module configured to determine an output value for the node based on the threshold values and the intermediate value in accordance with the sample transfer function.

6. The product of claim 5 wherein the output processing module comprises:

a comparing module configured to compare the intermediate value to the threshold values and to determine the output value for the node in accordance with the sample transfer function if the intermediate value is between the threshold values.

7. The apparatus according to claim 5 wherein the output processing module comprises:

a shifter configured to shift the intermediate value.