SYSTEM AND METHOD FOR COGNITIVE PROCESSING FOR DATA FUSION

• US 20120136913A1
• Filed: 03/07/2011
• Published: 05/31/2012
• Est. Priority Date: 03/15/2010
• Status: Active Grant

• Current Assignee: California Institute of Technology
• Original Assignee: California Institute of Technology
• Inventors: DUONG, Tuan A., Duong, Vu A.
• Granted Patent: US 8,204,927 B1
• Time in Patent Office: Days
• US Class Current: 708/819
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G11C 11/54 using elements simulating biological cells, e.g. neuron

Abstract

A system and method for cognitive processing of sensor data. A processor array receiving analog sensor data and having programmable interconnects, multiplication weights, and filters provides for adaptive learning in real-time. A static random access memory contains the programmable data for the processor array and the stored data is modified to provide for adaptive learning.

Claims

1. (canceled)

2. A system for processing sensor data comprising:

an input/output bus, wherein the input/output bus receives sensor data and outputs processed sensor data;

a processor array, wherein the processor array receives analog signals containing sensor data from the input/output bus and the processor array comprises one or more matrices of analog multiplication nodes and transfer function elements and the processor array outputs the processed sensor data from the one or more matrices of analog multiplication nodes and transfer function elements; and

a memory containing data values controlling configurations of the processor array,

wherein the processor array comprises;

a first array block, the first array block comprising;

a first programmable switch array configured to receive the sensor data;

a square array of first array analog multiplication nodes, wherein each first array multiplication node selectively receives analog signals from the first programmable switch array, or one or more first array multiplication nodes, or the first programmable switch array and one or more first array multiplication nodes; and

one or more first array transfer function elements, wherein each first array transfer function elements receives analog signals from one of the first array analog multiplication nodes, and

a second array block, the second array block comprising;

a second programmable switch array configured to receive signals from the one or more first array transfer function elements;

a cascaded array of second array analog multiplication nodes;

one or more second array transfer function elements, wherein each second array transfer function element receives analog signals from one of the second array analog multiplication nodes;

wherein each second array multiplication node selectively receives analog signals from the second programmable switch array, one or more second array multiplication nodes, the second programmable switch array and one or more second array multiplication nodes, or one of the one or more second array transfer function elements.

3. The system according to claim 2, wherein outputs from one or more of the first array transfer function elements are selectively coupled to the first programmable switch array as inputs to the first programmable switch array.

4. The system according to claim 2, wherein outputs from one or more of the second array transfer function elements, or outputs from one or more of the second array transfer function elements, or outputs from one or more of the second array transfer function elements and outputs from one or more of the second array transfer function elements are selectively coupled to the second programmable switch array as inputs to the second programmable switch array.

5. The system according to claim 3, wherein outputs from one or more of the second array transfer function elements, or outputs from one or more of the second array transfer function elements, or outputs from one or more of the second array transfer function elements and outputs from one or more of the second array transfer function elements are selectively coupled to the second programmable switch array as inputs to the second programmable switch array and selectively coupled to the first programmable switch array as inputs to the first programmable switch array.

6. The system according to claim 2, wherein the memory contains data values for selectively controlling the first programmable switch array or the second programmable switch array or the first programmable switch array and the second programmable switch array.

7. The system according to claim 2, wherein the memory contains data values for controlling multiplication factors for one or more of the first array multiplication nodes and second array multiplication nodes.

8. The system according to claim 7, wherein one or more of the first array multiplication nodes and second array multiplication nodes comprise a current mode multiplying digital to analog converter having a digital multiplication factor stored in the memory.

9. The system according to claim 2, wherein data values in the memory are modified based upon the processed sensor data.

10. The system according to claim 2, wherein the transfer function elements comprise programmable analog filters.

11. A method for processing sensor data comprising:

receiving sensor data in a plurality of first analog data streams;

applying a first set of multiplicative weights to the plurality of first analog data streams to produce a plurality of first multiplied analog data streams;

filtering the plurality of first multiplied analog data streams based on a first set of filter characteristics to produce a plurality of preprocessed analog data streams;

providing a plurality of second analog data streams based on the plurality of preprocessed analog data streams;

applying a second set of multiplicative weights to the plurality of second analog data streams to produce a second plurality of multiplied analog data streams;

filtering the second plurality of multiplied analog data streams based on a second set of filter characteristics to produce a plurality of processed analog data streams; and

outputting the plurality of processed analog data streams to produce processed sensor data.

12. The method according to claim 11, wherein the plurality of first analog data streams comprises:

sensor data, sensor data and the plurality of preprocessed analog data streams;

sensor data and the plurality of processed data streams;

or sensor data and the plurality of preprocessed analog data streams and the plurality of processed data streams.

13. The method according to claim 12, wherein the plurality of second analog data streams comprises:

the plurality of preprocessed analog data streams, or the plurality of preprocessed analog data streams and the plurality of processed analog data streams.

14. The method according to claim 13, wherein applying a first set of multiplicative weights comprises selectively directing the plurality of analog streams to a plurality of first multiplication nodes, wherein each first multiplication node multiplies a first multiplication node input by a weight selected from the first set of multiplicative weights to produce a first multiplication node output, wherein the first multiplication node input comprises:

one analog data steam of the plurality of analog data streams;

a first multiplication node output from another first multiplication node;

or a sum of one analog data steam of the plurality of analog data streams and a first multiplication node output from another first multiplication node, and wherein first multiplication node outputs from selected first multiplication nodes comprise the first plurality of multiplied analog data streams.

15. The method according to claim 13, wherein filtering the first plurality of multiplied analog data streams comprises:

providing an array of first analog filters, each first analog filter having a first analog filter input and a first analog filter output and a first analog filter characteristic selected from the selected first filter characteristics; and

,

directing each multiplied analog data stream of the first plurality of multiplied analog data streams into a first analog filter input of a selected first analog filter,

wherein first analog filter outputs comprise the plurality of preprocessed analog data streams.

16. The method according to claim 13, wherein applying a second set of multiplicative weights comprises:

selectively directing the plurality of preprocessed analog data streams to a plurality of second multiplication nodes, wherein each second multiplication node multiplies a second multiplication node input by a weight selected from the second set of multiplicative weights to produce a second multiplication node output, wherein the second multiplication node input comprises;

one preprocessed analog data steam of the plurality of preprocessed analog data streams;

a second multiplication node output from another second multiplication node;

or a sum of one preprocessed analog data steam of the plurality of preprocessed analog data streams and a second multiplication node output from another second multiplication node, and wherein second multiplication node outputs from selected second multiplication nodes comprise the second plurality of multiplied analog data streams.

17. The method according to claim 13, wherein filtering the second plurality of multiplied analog data streams comprises:

providing an array of second analog filters, each second analog filter having a second analog filter input and a second analog filter output and a second analog filter characteristic selected from the selected second filter characteristics; and

,

selectively directing each multiplied analog data stream of the second plurality of multiplied analog data streams to a second analog filter input of a selected second analog filter or to a selected third multiplication node of a plurality of third multiplication nodes, wherein each third multiplication node multiplies a third multiplication node input by a weight selected from a third set of multiplicative weights selected from the second filter characteristics to produce a third multiplication node output, wherein the third multiplication node input comprises;

a selected second analog filter output;

a third multiplication node output from another third multiplication node;

or a sum of a selected second analog filter output and a third multiplication node output from another third multiplication node; and

selectively directing each second analog filter output to a selected third multiplication node of a plurality of third multiplication nodes or to the plurality of processed analog data streams.

18. The method according to claim 11, further comprising

selecting or modifying at least one of the following sets based upon the processed sensor data:

the first set of multiplicative weights;

the first set of filter characteristics;

the second set of multiplicative weights; and

, the second set of filter characteristics.

19. The method according to claim 11, wherein the processed sensor data provides knowledge about a sensor environment and the method further comprises selecting or modifying at least one of the following sets based upon the sensor environment knowledge:

the first set of multiplicative weights;

the first set of filter characteristics;

the second set of multiplicative weights; and

, the second set of filter characteristics.

20. The method according to claim 11, further comprising converting sensor data received in a digital format to an analog format.