Data processing: artificial intelligence – Neural network – Structure
Reexamination Certificate
1998-10-09
2003-01-07
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Neural network
Structure
C706S034000
Reexamination Certificate
active
06505182
ABSTRACT:
CROSS-REFERENCES BY THE SAME AUTHOR
U.S. Pat. No. 4,809,222: Associative and Organic Memory and Methods. Filed Jun. 20, 1986, Issued Feb. 28, 1989.
U.S. Pat. No. 4,984,176: The VDH Biocomputer. Filed Jan. 11, 1988, Issued Jan. 8, 1991.
U.S. Pat. No. 5,375,250: Method of Intelligent Computing and Neural-Like Processing of Time and Space Functions. Filed Jul. 13, 1992, Issued Dec. 20, 1994.
U.S. Pat. No. 5,503,161: Universal Medical Instrument Based on Spectrum Analysis. Filed Oct. 25, 1993, Issued Apr. 2, 1996.
BACKGROUND
U.S. Pat. No. 4,809,222: Associative and Organic Memory and Methods, filed Jun. 20, 1986 and issued Feb. 28, 1989 documents an exploration of “intelligence” in computers. One of the references cited was an article entitled “Neural Networks are Naive, Says Minsky.” In 1986 the experts were pondering the potential of neurocomputing and the meaning of the word “intelligence.” The invention's contribution was a “memory that can forget.” This led to a neural network with surprising abilities, for instance changing priorities and real-time frequency spectrum analysis without the painful computations and memory requirements. Even though the proposed embodiment used analog circuits, the claims allowed for the use of digital circuits—for instance multipliers—“when they become available”.
U.S. Pat. No. 4,984,176: The VDH Biocomputer, filed Jan. 11, 1988 and issued Jan. 8, 1991, further explored the requirements of a computer for “intelligence.” The issue of software was also explored. The main idea was to minimize the added hardware and software complexity when the scale and scope of the applications increase.
U.S. Pat. No. 5,375,250: Method of Intelligent Computing and Neural-Like Processing of Time and Space Functions, filed Jul. 13, 1992 and issued Dec. 20, 1994, redefines the hardware used in the previous patent and further elaborates on the functions that can be performed in real time by the neural network first introduced in U.S. Pat. No. 4,809,222 (Associative and Organic Memory and Methods.) Algorithms—heuristic searches, back propagation of errors, etc.—that are used in more conventional neural networks apply but were not elaborated on. In this patent the neural engine was renamed “Resonant Processor.”
U.S. Pat. No. 5,503,161: Universal Medical Instrument Based on Spectrum Analysis, filed Oct. 25, 1993 and issued Apr. 2, 1996, shows how spectrum analysis—of light, sound and chemicals—can play a key role in devising a universal instrument for medicine, and how this instrument's general usefulness in medicine could be compared to that of the oscilloscope in electronics. In order to fully appreciate the timeliness of this patent, it is advisable not to skip the section entitled “Background of the Invention.” The citations cover a range of social and technical requirements. It promotes a paradigm where resources are concentrated on the practitioner at the bedside. This patent was written after the author had actually built a working version of his neural network and demonstrated that it actually works for real-time frequency spectrum analysis. By now the neural engine was variously referred to as a “resonant processor,” a biocomputer” or an “artificial cochlea” when programmed to work like the inner ear.
SUMMARY
The apparatus of the Claims that follow is the same as that used in the neural engine of the previously referenced patents by the same author, but with further modifications and/or additions. All versions make it possible to realize a neural network with neurons that are capable of storing a multivalued quantity, said multivalued quantity also being able to grow or decay as a function of time.
In a first aspect, whereas the previous patents did describe a general methodology for performing frequency spectrum analysis, the present invention provides a number of additional hardware and software enabling details.
In a second aspect, whereas the previous versions used two multipliers and one adder to perform arithmetic, the present invention replaces one of the multipliers with a Coincidence Detector (
3
), thus significantly expanding the scope and capabilities of the neural recognition engine in areas other than frequency spectrum analysis.
REFERENCES:
patent: 5140538 (1992-08-01), Bass et al.
patent: 5506915 (1996-04-01), Takstori et al.
patent: 5671336 (1997-09-01), Yoshida et al.
patent: 5875439 (1999-02-01), Engel et al.
Hikawa, H., “Learning Performance of Frequency-Modulation Digital Neural Network With On-Chip Learning”, IEEE Neural Network Proceedings on International Joint Conference on Computational Intelligence, May 1998.*
Chen et al, “Hybrid Architecture for Analogue Neural Network and Its Circuit Implementation”, IEE Proceedings on Circuits, Devices and Systems, Apr. 1996.*
Miura et al, “A Magnetic Neural Network, Utilizing Universal Arithmetic Modules for Pulse-Train Signal. Processing” IEEE Inter. Conf. on System Engineering, Sep. 1992.
LandOfFree
Recognition engine with time referenced neurons does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Recognition engine with time referenced neurons, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Recognition engine with time referenced neurons will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3051453