Parallel associative learning memory for a standalone hardwired recognition system
First Claim
Patent Images
1. A recognition system, comprising:
- a dynamically reconfigurable logic device similar to a field programmable gate array and programmed to do a feature extraction process from a media input stream; and
a zero instruction set computer (ZISC) device that includes a plurality of neural networks that are connected to simultaneously receive a preprocessed feature extraction data from the dynamically reconfigurable logic device;
wherein, a variety of recognition jobs can each be processed by the combination of the logic device and the ZISC by changing said feature extraction process via a dynamic reconfiguration and substantially no hardware changes.
1 Assignment
0 Petitions
Accused Products
Abstract
A recognition system comprises at least two field-programmable logic array devices connected to a common vector-input port of an array of a zero-instruction-set computers. Each field-programmable logic array device is configured to preprocess data from different respective media inputs and provide feature extraction vectors to the common vector-input port. Neural networks within the zero-instruction-set computer recognize the input patterns by comparing in parallel their vectors with those stored in each neural network cell. A variety of recognition jobs are made possible by changing the programming on-the-fly of the field-programmable logic array devices to suit each new job.
96 Citations
11 Claims
-
1. A recognition system, comprising:
-
a dynamically reconfigurable logic device similar to a field programmable gate array and programmed to do a feature extraction process from a media input stream; and
a zero instruction set computer (ZISC) device that includes a plurality of neural networks that are connected to simultaneously receive a preprocessed feature extraction data from the dynamically reconfigurable logic device;
wherein, a variety of recognition jobs can each be processed by the combination of the logic device and the ZISC by changing said feature extraction process via a dynamic reconfiguration and substantially no hardware changes. - View Dependent Claims (2, 3, 4, 5, 6, 7)
a plurality of dynamically reconfigurable logic devices each respectively programmed to do a unique feature extraction process from a corresponding and separate media input stream, and to output all extracted features onto a common bus; and
a plurality of zero instruction set computer (ZISC) devices each with a vector input connected to said common bus to simultaneously match all extracted features;
wherein the plurality of zero instruction set computer (ZISC) devices is expandable into a massively parallel configuration.
-
-
3. The system of claim 1, further including:
-
a memory in which said media input stream is stored; and
a region-of-interest (ROI) controller providing for the selection of a portion of said media input stream to be transferred from the memory to any of the dynamically reconfigurable logic devices.
-
-
4. The system of claim 3, further including:
a hypothesis generator connected to receive recognition results from the plurality of ZISC devices and providing bounds information to the ROI controller for particular regions-of-interest stored in the memory.
-
5. The system of claim 4, wherein:
the hypothesis generator constrains each recognition result by cross-correlating preliminary pattern recognitions with one another and eliminating conflicts before outputting a list of recognized items to a user.
-
6. The recognition system of claim 1, further comprising:
an image sensor connected to provide a video image to the dynamically reconfigurable logic device wherein realtime image capture and recognition processing can proceed without executing a software program.
-
7. The recognition system of claim 6, further comprising:
a removable memory storage device providing for storage of image files obtained by the image sensor, wherein a plurality of stored image files provide for system learning, verification, and on-site testing.
-
8. A recognition system, comprising:
-
at least two field-programmable logic array devices each separately configurable as a preprocessor for recognition feature extraction and connected to output a series of extracted-feature vectors to a common input neuron layer; and
a plurality of massively parallel zero-instruction-set computer devices each connected to receive said extracted-feature vectors from said common first neuron layer bus and each including a plurality of radial basic function (RBF) neural networks with a three layer structure for recognizing patterns within said extracted-feature vectors;
wherein, said three layer structure includes said common input neuron layer as its first layer, a second hidden layer formed of neuron circuits, and a neuron output layer that functions as the output of said neuron circuits. - View Dependent Claims (9, 10)
said RBF-neural networks operate to recognize any input patterns in said extracted-feature vectors by comparing in parallel such vectors with a plurality of vectors stored in each neural network cell.
-
-
10. The recognition system of claim 9, further comprising:
a dynamic reprogrammer connected to the field-programmable logic array devices that provides for a variety of recognition jobs to be made possible by changing any programming on-the-fly of the field-programmable logic array devices to suit each new job.
-
11. A recognition system, comprising:
-
a dynamically reconfigurable logic device similar to a field programmable gate array and programmed to do a feature extraction process from a media input stream; and
a zero instruction set computer (ZISC) device that includes a plurality of neural networks that are connected to simultaneously receive a preprocessed feature extraction data from the dynamically reconfigurable logic device;
wherein, a variety of recognition jobs can each be processed by the combination of the logic device and the ZISC by changing said feature extraction process via a dynamic reconfiguration and substantially no hardware changes and no computer instruction.
-
Specification