Methods and systems for neural and cognitive processing

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.

What is claimed is: 
     
         1 . A method of directed neural graph execution comprising managing execution of a plurality of neural function modules responsive to neural signals conveyed by a plurality of directed neural function module links that organize a portion of the plurality of neural function modules for execution as a directed neural graph. 
     
     
         2 . The method of  claim 1 , wherein managing execution of a plurality of neural function modules includes allocating a first portion of the plurality of neural function modules and corresponding directed neural function module links to processing nodes and communication paths respectively in a multi-processor computing system. 
     
     
         3 . The method of  claim 2 , wherein allocating a first portion of the plurality of neural function modules comprises allocating a portion of at least one neural function module to at least one processing node. 
     
     
         4 . The method of  claim 2 , wherein the allocating is responsive to execution of the directed neural graph. 
     
     
         5 . The method of  claim 2 , further comprising allocating a second portion of the plurality of neural function modules to processing modules in the multi-processor computing system responsive to execution of the first portion. 
     
     
         6 . The method of  claim 1 , wherein managing execution of the plurality of neural function modules comprises maintaining a non-decreasing value virtual clock that facilitates synchronized execution of distinct portions of the directed neural graph. 
     
     
         7 . The method of  claim 1 , wherein managing execution of a plurality of neural function modules comprises executing the directed neural graph on one or more of physical computer processors and virtual computer processors. 
     
     
         8 . The system of  claim 7 , wherein managing execution of a plurality of neural function modules involves multiprocessor coordination. 
     
     
         9 . A method of neural process execution comprising executing a directed graph representation of a neural process specification by managing a plurality of neural function modules connected in the specification through a plurality of links that convey multiplexed neural signals between at least two of the plurality of neural function modules. 
     
     
         10 . A system comprising:
 a plurality of neural function modules stored in a non-volatile processor accessible memory;   
       a plurality of directed neural function module links disposed to organize portions of the plurality of neural function modules in the memory for execution as a directed neural graph; and
 a neural graph execution facility executing on at least one processor, the facility managing execution of the plurality of neural function modules responsive to neural signals conveyed by the plurality of directed neural function module links. 
 
     
     
         11 . The system of  claim 10 , wherein the neural graph execution facility is operable by a multi-processor computing system, the facility operable based on an allocation of a portion of the plurality of neural function modules to processing nodes in the multi-processor computing system. 
     
     
         12 . The system of  claim 10 , wherein portions of the directed neural graph that do not share at least one of the plurality of directed neural function module links are executed independently per an allocation of the neural function modules to processing nodes in a multi-node processor computing system. 
     
     
         13 . The system of  claim 10 , wherein the neural graph execution facility further allocates each of a portion of the plurality of neural function modules to at least one processing node in a multi-processor computing system, the allocating based on the directed neural function module link-based organization of the neural function modules. 
     
     
         14 . The system of  claim 10 , further comprising a non-decreasing value virtual clock that facilitates synchronized execution of distinct portions of the directed neural graph. 
     
     
         15 . The system of  claim 10 , wherein the at least one processor comprises one or more of physical computer processors and virtual computer processors. 
     
     
         16 . The system of  claim 10 , wherein the neural graph execution facility manages execution of a plurality of neural function modules by coordinating operation of a plurality of processors for operating the neural graph execution facility. 
     
     
         17 . The system of  claim 10 , wherein the neural graph execution facility is adapted to share a plurality of pattern memory spaces among the plurality of neural function modules. 
     
     
         18 . The system of  claim 10 , wherein the neural graph execution facility further manages access by the neural function modules to a plurality of shared pattern memory spaces. 
     
     
         19 . The system of  claim 10 , wherein the neural graph execution facility comprises a compilation function, an executable process representation, and an execution machine, where the compilation function generates the executable process representation corresponding to the directed neural function module link-based organization of the neural function modules, and the execution machine executes said executable process representation. 
     
     
         20 . The system of  claim 10 , wherein the neural graph execution facility comprises an interpreter function and an execution module, wherein the interpreter function provides an executable representation of an interpreted portion of the directed neural graph that is executed by the execution module.