Dynamical brain model for use in data processing applications

• US 7,398,256 B2
• Filed: 02/25/2002
• Issued: 07/08/2008
• Est. Priority Date: 02/23/2001
• Status: Active Grant

• Current Assignee: Pennsylvania University of The Trustees of The
• Original Assignee: Trustees Of The University Of Pennsylvania (University of Pennsylvania)
• Inventors: Farhat, Nabil H
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US10/468,135
• Publication Number:
  US 20040073415A1
• Time in Patent Office: 2,325 Days
• Field of Search: None
• US Class Current: 706/12
• CPC Class Codes: G06N 3/02 Neural networks

Abstract

A system and methods offering a dynamical model of cortical behavior is provided. In an illustrative implementation, the present invention offers a corticonic network comprising at least one parametrically coupled logistic map network (PCLMN)(205). The PCLMN offers a non-linear iterative map of cortical modules (or netlets) that when executed exhibit substantial cortical behaviors. The PCLMN accepts dynamic and/or static spatio-temporal input (210) and determines a fixed point attractor in state-space for that input. The PCLM (205) operates such that if the same or similar dynamic and/or static spatio-temporal input is offered over several iterations, the PCLMN converges to the same fixed point attractor is provided rendering adaptive learning. Further, the present invention contemplates the memorization or association of inputs using the corticonic network in a configuration where the PCLMN cooperates with another cortical module model (e.g. another PCLMN, associative memory module, etc.)(215).

Claims

1. A method to simulate cortical processing for use in processing data, comprising:

segmenting a cortex into cortical columns, said cortical columns communicating with each other via short-range and longer-range communications paths;

generating at least one parametrically coupled logistic map network (PCLMN) to model these communication paths, said PCLMN comprising one or more parametrically coupled logistic maps;

using a difference equation in the PCLMN that represents neurons in the cortical columns in the modeling of the cortex; and

configuring the PCLMN to perform acts comprising any of (a) handling dynamic and/or static input patterns, (b) handling large numbers of sensory stimuli, and (c) providing a solution to a plasticity-stability problem to memory formation.

9. A computer readable storage medium having computer readable instructions to perform the steps recited in claim 1.

2. A method to simulate cortical processing for use in processing data, comprising:

segmenting a cortex into cortical columns, said cortical columns communicating with each other via short-range and longer-range communications paths;

generating at least one parametrically coupled logistic map network (PCLMN) to model these communication paths, said PCLMN comprising one or more parametrically coupled logistic maps;

using a difference equation in the PCLMN that represents neurons in the cortical columns in the modeling of the cortex;

providing at least one dynamic and/or static input to said at least one PCLMN; and

processing said at least one dynamic and/or static input by said PCLMN to determine a fixed-point attractor for said at least one dynamic and/or static input in state-space.

3. The method as recited in claim 2, further comprising:

offering said at least said attractor by said PCLMN for similar or same said at least one dynamic and/or static input, said PCLMN engaging in adaptive learning to offer said attractor.

4. The method as recited in claim 3, further comprising:

providing at least one additional cortical model module to cooperate with said at least one PCLMN to perform memorization and/or association operations, said at least one additional cortical module comprising any of a PCLMN and associative memory.

5. The method as recited in claim 4, further comprising:

providing a feedback loop to allow for the design and implementation of desired state-space trajectories, said feedback loop originating from the output of said at least one additional cortical model module and terminating at the input of said at least one PCLMN.

6. The method as recited in claim 5 further comprising:

labeling said determined attractor by said at least one additional cortical model module for association and storage.

7. The method as recited in claim 3, wherein said at least dynamic and/or static inputs comprise dynamic and/or static spatio-temporal stimuli.

8. The method as recited in claim 7, wherein said dynamic and/or static spatio-temporal stimuli originate from at least one filter, said filter comprising any of an electronic filter, a partially-electronic filter, and a biological filter.

10. A system to model cortical processing for use in processing data, comprising:

a memory;

at least one parametrically coupled logistic map network (PCLMN) stored in the memory, said PCLMN providing a dynamic and/or static non-linear representation of cortical connections as observed in a biological cortex using a difference equation that represents neurons in cortical columns,

wherein said at least one PCLMN accepts at least one dynamic and/or static input and processes said dynamic and/or static input to identify a fixed-point attractor for said at least one dynamic and/or static input in state space,

wherein said at least one PCLMN operates to adapt providing the same output to similar or same inputs thereby engaging in learning; and

wherein said at least one dynamic and/or static input comprises at least one spatio-temporal stimulus.

11. A system to model cortical processing for use in processing data, comprising:

a memory;

at least one parametrically coupled logistic map network (PCLMN) stored in the memory, said PCLMN providing a dynamic and/or static non-linear representation of cortical connections as observed in a biological cortex using a difference equation that represents neurons in cortical columns,

wherein said at least one PCLMN accepts at least one dynamic and/or static input and processes said dynamic and/or static input to identify a fixed-point attractor for said at least one dynamic and/or static input in state space,

wherein said at least one PCLMN operates to adapt providing the same output to similar or same inputs thereby engaging in learning; and

wherein said adaptation of said at least one PCLMN occurs through few iterations of inputs.

12. A system to model cortical processing for use in processing data, comprising:

a memory;

at least one parametrically coupled logistic map network (PCLMN) stored in the memory, said PCLMN providing a dynamic and/or static non-linear representation of cortical connections as observed in a biological cortex using a difference equation that represents neurons in cortical columns,

wherein said at least one PCLMN accepts at least one dynamic and/or static input and processes said dynamic and/or static input to identify a fixed-point attractor for said at least one dynamic and/or static input in state space,

wherein said at least one PCLMN operates to adapt providing the same output to similar or same inputs thereby engaging in learning; and

at least one additional cortical model module, said at least one additional cortical model module accepting the output of said at least one PCLMN as input and is capable of classifying in real time said output of said at least one PCLMN as being novel simulating memorization and/or association cortical functions.

13. The system as recited in claim 12, wherein said at least one additional cortical model module comprises any of a PCLMN and associative memory.

14. The system as recited in claim 13, wherein said at least one PCLMN and said at least one additional cortical model module are tandemly coupled.

15. The system as recited in claim 14, wherein said at least one PCLMN and said at least one additional cortical model operate synchronously.

16. The system as recited in claim 14 further comprising:

a feedback loop originating from the output of said at least one additional cortical model module and terminating at the input of said at least one PCLMN, wherein said feed back loop allowing for the design and implementation of desired state-space trajectories.

17. A method to model the cortex and use said model in order to process input data, comprising:

providing a corticonic network, said corticonic network comprising at least one parametrically coupled logistic map network (PCLMN) operating on at least one dynamic and/or static input, said at least one dynamic and/or static input comprising spatio-temporal stimulus and said PCLMN modeling cortical functions using a difference equation;

processing said at least one dynamic and/or static input using said corticonic network;

providing a fixed point attractor in state space for said at least one dynamic and/or static input; and

performing memorization and/or association on said outputs of said corticonic network, said corticonic network engaging in memorization through the use of at least one additional PCLMN tandemly coupled to said at least one PCLMN and operating to classify novel output data from said at least one PCLMN in order to process the input data.

23. The method as recited in claim 17, further comprising performing self-organization, wherein the self-organization comprises undertaking driven adaptation such that inputs to the corticonic network are rapidly classified using input-specific attractors.

24. The method as recited in claim 17, further comprising employing chaos principles to perform the driven adaptation.

25. The method as recited in claim 17, further comprising distinguishing between structured and random input patterns.

26. The method as recited in claim 17, further comprising ascertaining coexisting attractors that are uniquely accessed by the inputs.

27. The method as recited in claim 26, further comprising employing an associative memory network to label the attractors, wherein the associative memory network comprises a bank of -D optical holographic associative memories.

28. The method as recited in claim 17, further comprising labeling inputs with time specific attractors, wherein the attractor is exists for a set period of time.

29. The method as recited in claim 17, further comprising performing symmetry breaking, wherein the symmetry breaking is realized by randomly selecting a fraction of the processing elements and isolating the elements from their neighbors.

30. The method as recited in claim 29, further comprising partitioning the corticonic network into sub-networks of unequal size that are isolated from each other.

31. The method as recited in claim 17, further comprising inputting invariant feature-vector(s) of an object to the corticonic network.

18. A system modeling the cortex for use in processing data comprising:

a memory;

a corticonic network,

wherein said corticonic network comprises a first parametrically coupled logistic map network (PCLMN) stored in said memory, said first PCLMN accepting dynamic and/or static inputs and determining a fixed point attractor in state-space to describe at least one dynamic and/or static input, said first PCLMN models close-range intercortical connections by space-time filtering of the inputs in order to reduce redundancies in patterns of the inputs using an autonomous adaptation algorithm, and

wherein said corticonic network comprises a second PCLMN, said second PCLMN tandemly coupled to said first PCLMN to accept the output of said first PCLMN as input and operates to identify non-novel output of said first PCLMN, said identification of non-novel output representative of memorization and/or association cortical functions, and

wherein a feedback ioop operates between the output of said second PCLMN and the input of said first PCLMN.

19. The system as recited in claim 18, wherein said corticonic network comprises a computing environment.

20. The method as recited in claim 18, wherein said second PCLMN exhibits associative memory attributes to label and associate attractors resulting from said first PCLMN.

32. The system as recited in claim 18, further comprising including a feedback from a labeling associated memory to the first PCLMN such that a cortical module is produces which is capable of producing discrete-time trajectories in the state-space of the cortical module.

33. The system as recited in claim 32, wherein the trajectories can be fashioned to provide the cortical module with categorical-perception and perceptual inference.

21. A method of applying a dynamical brain model for processing data comprising:

identifying a data processing problem;

modeling the data processing problem in a corticonic network;

configuring said corticonic network to the parameters of said data processing problem;

executing said corticonic network using a computer, the corticonic network modeled by a parametrically coupled logistic map (PCLM) that models cortical functions using a difference equation;

using the corticonic network for processing the data and providing a processed output;

obtaining results from said execution of said corticonic network; and

verifying said results of said execution of said corticonic network by comparing said results of the processed output with a comparable real-world model.

22. A computer readable storage medium comprising computer readable instructions to perform the steps recited in claim 21 in order to process the input data.

34. The method as recited in claim 21, further comprising processing spatio-temporal and static input patterns.