Methods and systems for neural and cognitive processing

• US 10,417,554 B2
• Filed: 05/21/2015
• Issued: 09/17/2019
• Est. Priority Date: 05/22/2014
• Status: Active Grant

• Current Assignee: Lee J. Scheffler
• Original Assignee: Lee J. Scheffler
• Inventors: Scheffler, Lee J.
• Primary Examiner(s): Vincent, David R
• Application Number:
  US14/718,642
• Publication Number:
  US 20150339570A1
• Time in Patent Office: 1,580 Days
• Field of Search: 706 12, 706 15, 706 45
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/10 Interfaces, programming languages or software development kits, e.g. for simulating neural networks

Abstract

Provided herein is a system for creating, modifying, deploying and running intelligent systems by combining and customizing the function and operation of reusable component modules arranged into neural processing graphs which direct the flow of signals among the modules, analogous in part to biological brain structure and operation as compositions of variations on functional components and subassemblies.

Claims

1. A neural process specification generation and execution system comprising a neural process specification configuration environment and a neural computing environment, said neural process specification configuration environment further comprising an interface by which a plurality of neural function modules and a plurality of links for conveying a plurality of multiplexed neural signals there between are configured into a neural process specification of an executable directed graph, said neural computing environment executing said directed graph by managing at least one of said plurality of neural function modules of said neural process specification responsive to a change in at least one of said plurality of multiplexed neural signal conveyed over at least one of said plurality of links to at least one of said plurality of neural function modules.

2. The system of claim 1, wherein said neural process specification further comprises one or more pattern memory spaces, each said memory pattern space holding one or more memory patterns, each of said one or more memory patterns representing a combination of neural signal values, and each said memory space being sharable by one or more said neural function modules in the neural process specification.

3. The system of claim 2, wherein at least one neural function module is a memory pattern module accessing one or more memory patterns in one or more said memory pattern spaces.

4. The system of claim 3, wherein said memory pattern module compares a combination of input signals comprising a portion of the plurality of multiplexed neural signals with a memory pattern in said one or more memory patterns, deriving for said memory pattern a pattern matching score reflecting a degree of matching of said input signal combination with said memory pattern and sends an output signal for said pattern using a function of said pattern matching score as the signal value.

5. The system of claim 4, wherein said function is one of a constant function, a binary function, a linear function, a logistic function or mathematical combinations thereof.

6. The system of claim 3, wherein said neural function module additionally creates a new memory pattern corresponding to an input signal combination that fails to match any existing said memory pattern with a matching score above a threshold value.

7. The system of claim 3, wherein said memory pattern module additionally makes pattern adjustments to the one or more memory patterns to incorporate an input signal combination, a magnitude of said pattern adjustments governed by a learning rate.

8. The system of claim 7, wherein said pattern adjustments include one of changing a weight associated with an input signal, changing an expected value associated with an input signal, adding a new signal to the memory pattern, and removing a signal from the memory pattern.

9. The system of claim 4, wherein said neural function module is at least one of a set module, a sequence module or a temporal module.

10. The system of claim 9, wherein a said neural function module is selected from a group consisting of a weight set module and a weight value set module.

11. The system of claim 9, wherein a said neural function module is selected from a group consisting of a weight matrix sequence pattern module, a weight value matrix sequence pattern module, a regular expression sequence pattern module, an edit distance sequence pattern module, an open bigrams sequence pattern module, and a state machine sequence pattern module.

12. The system of claim 4, wherein a said neural function module is a temporal module.

13. The system of claim 2, wherein a said neural function module is a reify module generating an exemplar of a pattern, said exemplar comprising output signals corresponding to identifiers in said pattern, said generating responsive to an identity of said pattern that is received as an input signal by said reify module, said output signals representing at least one identifier that is different from said identity of said pattern.

14. A system for configuring and executing a neural process specification comprising:

a plurality of neural function modules, each neural function module comprising at least one of a neural signal link input and a neural signal link output;

a neural process specification interface by which a user configures an executable instance of at least a portion of the plurality of neural function modules, neural signal links that convey a plurality of neural signals from at least one neural signal link output of a first neural function module to at least one neural signal link input of a second neural function module, the configuring forming an executable directed graph; and

a neural process specification execution system that executes the directed graph by managing execution of the first neural function module and the second neural function module responsive to a change of a neural signal conveyed from the first neural function model to the second neural function module.

15. The system of claim 14, wherein at least one of the first neural function module and the second neural function module comprises a memory pattern adjusting module that, when executed by the neural process specification execution system adjusts at least one memory pattern, a magnitude of the adjusting governed by the learning rate.

16. The system of claim 15, wherein the neural process specification interface further facilitates configuring the neural signal links with a plurality of multiplexed neural signals that each are associated with a unique signal identifier that the neural process specification execution system uses to manage execution of the neural function modules in the directed graph.

17. The system of claim 16, wherein the neural process specification execution system activates a receiving neural function module to which a neural signal that corresponds to a specific unique signal identifier is conveyed responsive to a change in the neural signal that corresponds to the specific unique signal identifier.

18. The system of claim 14, wherein the neural process specification further includes a learning rate that applies to a neural function module that performs a machine learning function.

19. The system of claim 14, wherein the neural process specification further includes a learning rate that applies to a memory pattern adjusting neural function module that receives input from a machine learning neural function module.