Directed behavior in hierarchical temporal memory based system

• US 9,424,512 B2
• Filed: 01/07/2015
• Issued: 08/23/2016
• Est. Priority Date: 02/10/2006
• Status: Active Grant

• Current Assignee: Numenta, Inc.
• Original Assignee: Numenta, Inc.
• Inventors: Hawkins, Jeffrey C., George, Dileep
• Primary Examiner(s): CHEN, ALAN S
• Application Number:
  US14/591,821
• Publication Number:
  US 20150142710A1
• Time in Patent Office: 594 Days
• US Class Current: 1/1
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A hierarchy of computing modules is configured to learn a cause of input data sensed over space and time, and is further configured to determine a cause of novel sensed input data dependent on the learned cause. At least one of the computing modules has a sequence learner module configured to associate sequences of input data received by the computing module to a set of causes previously learned in the hierarchy.

Claims

1. An apparatus comprising:

a control module configured to store a plurality of unit behaviors associated with controlling movements of one or more effectors; and

a learning system coupled to the control module and comprises a plurality of nodes arranged into first level nodes and second level nodes, each of the first level nodes associated with a first level effector motion, each of the second level nodes associated with a second level effector motion, one or more first level effector motions combined to form a second level effector motion, the learning system configured to;

receive sensed inputs corresponding to movements of one or more effectors that occur responsive to invoking one or more of the plurality of unit behaviors at the control module; and

generate a representation of relationships between the invoked one or more unit behaviors and the one or more effectors based on the sensed inputs and the invoked one or more unit behaviors.

2. The apparatus of claim 1, wherein the learning system is further configured to make a prediction based on the generated representation of the relationships to cause an intended behavior by controlling the one or more effectors.

3. The apparatus of claim 1, wherein a node of the plurality of nodes is switched off to prevent first or second level effector motions corresponding to the node.

4. The apparatus of claim 1, wherein a signal from a node of the plurality of nodes is modified to change first or second level effector motions corresponding to the node.

5. The apparatus of claim 1, wherein at least a subset of the one or more effectors control capturing of images, and the sensed inputs represent the captured images.

6. The apparatus of claim 1, wherein the one or more effectors comprise at least a motor and the control module comprises processing circuitry for operating the motor.

7. The apparatus of claim 1, where the relationships pair at least a unit behavior with at least a subset of the effectors that causes the unit behavior.

8. The apparatus of claim 1, wherein the learning system comprise Hierarchical Memory System.

9. A method of controlling one or more effectors, comprising:

causing movements of one or more effectors by invoking one or more unit behaviors stored in a control module;

receiving, by a learning module, sensed inputs corresponding to the movements of the one or more effectors that occur responsive to invoking the one or more unit behaviors at the control module;

generating, by the learning module, a representation of relationships between the invoked unit behavior and the one or more effectors based on the sensed inputs and the one or more invoked unit behaviors;

associating each of first level nodes in a plurality of nodes in the learning module with a first level effector motion; and

associating each of second level nodes in the plurality of nodes with a second level effector motion, one or more first level effector motions combined to form a second level effector motion.

10. The method of claim 9, further comprising making prediction based on the representation of the relationships to cause an intended behavior by controlling the one or more effectors.

11. The method of claim 9, further comprising switching off a node of the plurality of nodes to prevent first or second level effector motions corresponding to the node.

12. The method of claim 11, further comprising modifying a signal from a node of the plurality of nodes to change second level effector motions corresponding to the node.

13. The method of claim 9, wherein at least a subset of the one or more effectors controls capturing of images, and the sensed inputs represent the captured images.

14. The method of claim 9, wherein the one or more effectors comprise at least a motor and the control module comprises processing circuitry for operating the motor.

15. The method of claim 9, where the relationships pair at least a unit behavior with at least a subset of the effectors that causes the unit behavior.

16. The method of claim 9, wherein the learning module comprise Hierarchical Temporal Memory module.

17. A non-transitory computer-readable storage medium storing instructions thereon, the instructions when executed by a processor cause the processor to:

cause movements of one or more effectors by invoking one or more unit behaviors stored in a control module;

receive, by a learning module, sensed inputs corresponding to the movements of the one or more effectors that occur responsive to invoking one or more of the unit behaviors at the control module;

generate, by the learning module, a representation of relationships between the invoked unit behavior and the one or more effectors based on the sensed inputs and the invoked unit behaviors;

associate each of first level nodes in a plurality of nodes in the learning module with a first level effector motion; and

associate each of second level nodes in the plurality of nodes with a second level effector motion, one or more first level effector motions combined to form a second level effector motion.

18. The non-transitory computer-readable storage medium of claim 17, further comprising instructions to make prediction based on the representation of the relationships to cause an intended behavior by controlling the one or more effectors.