Neural network that incorporates direct optical imaging

• US 5,842,191 A
• Filed: 09/03/1997
• Issued: 11/24/1998
• Est. Priority Date: 12/12/1994
• Status: Expired due to Term

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Stearns, Richard G.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Shah, Sanjiv
• Application Number:
  US08/922,798
• Time in Patent Office: 447 Days
• Field of Search: 395/25, 706/40
• US Class Current: 706/40
• CPC Class Codes: G06N 3/0675 using electro-optical, acousto-optical or opto-electronic means
• CPC Class Codes: G06V 10/20 Image preprocessing

Abstract

A compact neural network architecture is trainable to sense and classify an optical image directly projected onto it. The system is based upon the combination of a two-dimensional amorphous silicon photoconductor array and a liquid-crystal spatial light modulator. Appropriate filtering of the incident optical image upon capture is incorporated into the net work training rules, through a modification of the standard backpropagation training algorithm. Training of the network on two image classification problems is described: the recognition of handprinted digits, and facial recognition. The network, once trained is capable of standalone operation, sensing an incident image and outputting a final classification signal in real time.

Claims

1. An optically-addressed neural network comprising:

a two-dimensional array of photo conductors defining an array of M by N array nodes, wherein M and N are defined by positive integers and wherein the nodes of each column share a common column conductor and nodes of each row share a common row conductor;

a light mask disposed between the two-dimensional array and an associated light source such that generally collimated light therefrom is incident on a selected portion of the plurality of array nodes;

a trainable input portion of the array of photoconductors adapted to directly receive an optical image thereon, each common row conductor of the input portion having a voltage thereon fixed a preselected level;

a plurality of amplifiers, each operatively connected to one of (a) a selected one of the common column conductors of the array and (b) a selected one of the common row conductors of the array;

a first plurality of conductors for impressing selected amplified signals, received from the amplifiers, to a selected array feedback portion of the common row conductors unique to those common row conductors of the input portion; and

a second plurality of conductors, unique to the first plurality of conductors, forming a system output from selected amplified signals received from the amplifiers.

2. The optically-addressed neural network of claim 1 wherein the second plurality of conductors share no amplifier associated with any common column conductor of the input portion.

3. The optically-addressed neural network of claim 2 wherein the amplifiers are comprised of non-linear current-to-voltage converters.

4. The optically-addressed neural network of claim 3 wherein the light mask directs the generally collimated light of the associated light source to a first subset of the array elements defined as sharing no common row conductor with the input portion.

5. The optically-addressed neural network of claim 4 wherein the light mask directs the generally collimated light of the associated light source to the first subset of the array elements defined as sharing at least a portion of common row conductors associated with the system output.

6. The optically-addressed neural network of claim 5 wherein the first subset is rectangular.

7. The optically-addressed neural network of claim 5 wherein the first subset is comprised of first and second sub-portions sharing no common row conductor and no common column conductor therebetween.

8. The optically-addressed neural network of claim 5 wherein the first subset is comprised of first and second sub-portions sharing no common row conductor therebetween.

9. The optically-addressed neural network of claim 5 wherein the first subset is comprised of first and second sub-portions sharing no common column conductor therebetween.

10. The optically-addressed neural network of claim 9 further comprising a second subset of array elements sharing at least one of no column conductor and no common column conductor with one of the first and second sub-portions.

11. The optically-addressed neural network according to claim 1 wherein a single power supply supplies the voltage at a fixed preselected level.

12. The optically-addressed neural network according to claim 1 wherein intensity of the optical image varies spatially.

13. The optically-addressed neural network of claim 1, wherein at least one of each of the common row and common column conductors are configured to sum currents from photoconductors associated with the at least one of each of the common row and common column conductors.

14. The optically-addressed neural network of claim 1 wherein the data generated from the input of the optical image is compressed.

15. An optically-addressed neural network comprising:

a two-dimensional array of two-terminal photoconductors disposed in an array of M rows by N columns, wherein M and N are defined as positive integers, wherein

a first terminal of each photoconductor in each row M of N photoconductors is connected in common to a row M conductor, and

a second terminal of each photoconductor in each column N of M photo conductors is connected in common to a column N conductor;

a two-dimensional mask, generally parallel to the array, disposed between the array and an associated, relatively collimated light source, the mask including,

a plurality of selected subareas, each being set to a selected translucence level;

a plurality of J adjacent input conductors, each being electrically coupled to a corresponding one of the row conductors, wherein J is an integer less than M;

a network of amplifiers electrically coupled to each column conductor such that N amplified column signals are generated thereby;

an electrical coupling between each of a first subset of the N amplified column signals and corresponding adjacent ones of the row conductors unique to the J row conductors; and

an output formed from a second subset of the N amplified column signals unique to the first subset thereof;

the selected subareas of the mask including, a first window portion defining light passing therethrough so as to affect a subset w_kj of the photoconductors, which subset is defined by a subset of the columns of the photoconductors in the J rows,

a second window portion defining light passing therethrough so as to affect a subset w_lk of the photoconductors, which subset is defined by a subset of the columns of the photoconductors not disposed in the J rows, and

a third window portion so as to communicate the optical image to an area w_ji of the photoconductors, which area shares no column conductor or row conductor with the conductors within w_kj or w_lk, and wherein column conductors associated with photoconductors of the area w_ji comprise the J input conductors.

16. The optically-addressed neural network of claim 15 wherein the mask includes a generally opaque portion so as to prevent light from substantially affecting the photoconductors on all areas of the two-dimensional array other than w_kj and w_lk.

17. The optically-addressed neural network of claim 16 wherein photoconductors of the areas w_kj and w_lk share no column conductor and no row conductor.

18. The optically-addressed neural network of claim 15, wherein at least one of each of the common row and common column conductors are configured to sum currents from photoconductors associated with the at least one of each of the common row and common column conductors.

19. The optically-addressed neural network of claim 15 wherein the selected translucence level is obtained by training the neural network configured such that information is input between nodes rather than at nodes.

20. The optically-addressed neural network of claim 15 wherein the data generated from the input of the optical image is compressed.

27. The optically-addressed neural network of claim 15 wherein the second subset is formed from amplified column signals associated with one-half the number of column conductors of w_lk.

21. A neural method comprising the steps of:

transmitting an optical image, via a light mask disposed between a two-dimensional array of photoconductors defining a plurality of array nodes and a source of the optical image, to a selected area of the two-dimensional array, whereby the optical image is directly input into a trainable neural network, the selected area of the photoconductor array being selected in accordance with a training of the neural network, the training including inputting information between the array nodes rather than at the array nodes;

amplifying signals which are a function of the optical image input to the array nodes, with a plurality of amplifiers;

impressing at least some of the amplified signals, received from the amplifiers, to an array feedback portion of the array nodes; and

generating an output signal formed as a result of the received optical image.

22. The method of claim 21 wherein the step of transmitting the optical image to the selected area includes a step of selectively maintaining subportions of the mask at a substantially opaque transmissivity.

23. The method of claim 21 wherein the step of generating the output signal includes outputting a portion of the amplified signals, as the output signal.

24. The method of claim 21 wherein the step of impressing includes the step of impressing the amplified signals to the array feedback portion defined as unique to the first defined subset.

25. The method of claim 21 further comprising a step of compressing data generated from the input of the optical image.

26. An optically-addressed neural network comprising a two-dimensional array of photoconductors defining an array of M by N array nodes, wherein M and N are defined by positive integers and wherein the nodes of each column share a common column conductor and nodes of each row share a common row conductor;

an input portion of the photoconductor array adapted to directly receive an optical image thereon, the photoconductor array including a weight field adapted to receive input data corresponding to the optical image, thereby allowing training of the input portion;

a subset of the photoconductor array defined as sharing no common row conductor with the input portion, the subset comprised of first and second sub-portions sharing at least one of no common column conductor and no common row conductor;

a light mask disposed between the two-dimensional array and an associated light source such that generally collimated light therefrom is incident on the subset of the photoconductor array;

a plurality of amplifiers, each operatively connected to one of the common column conductors of the photoconductor array;

a first plurality of conductors for impressing selected amplified signals, received from the amplifiers, to a selected array feedback portion of the common row conductors unique to those common row conductors of the input portion; and

a second plurality of conductors, unique to the first plurality of conductors, forming a system output from selected amplified signals received from the amplifiers.