Sequence learning in a hierarchical temporal memory based system

• US 8,666,917 B2
• Filed: 09/05/2012
• Issued: 03/04/2014
• Est. Priority Date: 02/10/2006
• Status: Active Grant

• Current Assignee: Numenta, Inc.
• Original Assignee: Numenta, Inc.
• Inventors: Jaros, Robert G., George, Dileep, Hawkins, Jeffrey, Astier, Frank
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): Chubb, Mikayla
• Application Number:
  US13/604,543
• Publication Number:
  US 20120330885A1
• Time in Patent Office: 545 Days
• Field of Search: None
• US Class Current: 706/12
• 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. A computer-implemented system, comprising:

a processor;

a coincidence detector configured to;

detect spatial patterns in an input signal, the spatial patterns in the input signal changing over time; and

generate a series of coincidence detector outputs indicating the detected spatial patterns in the input signal; and

a sequence learner coupled to the coincidence detector and configured to;

associate temporal sequences of the detected spatial patterns by storing temporal relationships between the series of coincidence detector outputs;

detecting a temporal sequence associated with two or more adjacent and distinct spatial patterns detected by the coincidence detector based on the stored temporal relationships between the series of coincidence detector outputs; and

generate a sequence learner output indicating the detected temporal sequence.

2. The system of claim 1, wherein the sequence learner is configured to store the temporal relationships between the series of coincidence detector outputs in a list.

3. The system of claim 1, wherein the sequence learner output comprises numbers representing likelihood that the input signal is caused by objects or states of the objects.

4. The system of claim 1, wherein the coincidence detector and the sequence learner are implemented as software.

5. The system of claim 1, wherein the sequence learner, in a learning stage, is configured to remove a first sequence of spatial patterns responsive to receiving a second sequence of spatial patterns that is similar to the first sequence of spatial patterns.

6. The system of claim 1, wherein the sequence learner, in a learning stage, is configured to remove a sequence of spatial patterns based on occurrence frequency of the sequence of spatial patterns.

7. The system of claim 1, wherein the coincidence detector and the sequence learner are part of a computing module that forms a hierarchical structure with other computing modules.

8. A computer-implemented method, comprising:

at a coincidence detector, detecting spatial patterns in an input signal, the spatial patterns in the input signal changing over time;

at the coincidence detector, generating a series of coincidence detector outputs indicating the detected spatial patterns in the input signal;

at a sequence learner, associating temporal sequences of the detected spatial patterns by storing temporal relationships between the series of coincidence detector outputs;

at the sequence learner, detecting a temporal sequence associated with two or more adjacent and distinct spatial patterns detected by the coincidence based on the stored temporal relationships between the series of coincidence detector outputs; and

at the sequence learner, generating a sequence learner output indicating the detected temporal sequence.

9. The method of claim 8, wherein the temporal relationships between the series of coincidence detector outputs are stored in a list.

10. The method of claim 8, wherein the sequence learner output comprises numbers representing likelihood that the input signal is caused by objects or states of the objects.

11. The method of claim 8, wherein the coincidence detector and the sequence learner are implemented as software.

12. The method of claim 8, further comprising removing a first sequence of spatial patterns responsive to receiving a second sequence of spatial patterns that is similar to the first sequence of spatial patterns at the sequence learner in a learning stage.

13. The method of claim 8, further comprising removing a sequence of spatial patterns based on occurrence frequency of the sequence of spatial patterns at the sequence learner in a learning stage.

14. The method of claim 8, wherein the coincidence detector and the sequence learner are part of a computing module that forms a hierarchical structure with other computing modules.

15. A non-transitory computer-readable storage medium having instructions stored therein, the instructions when executed on a processor cause the processor to:

detect spatial patterns in an input signal, the spatial patterns in the input signal changing over time;

generate a series of coincidence detector outputs indicating the detected spatial patterns in the input signal;

associate temporal sequences of the detected spatial patterns by storing temporal relationships between the series of coincidence detector outputs;

detect a temporal sequence associated with two or more adjacent and distinct spatial patterns that are detected based on the stored temporal relationships between the series of coincidence detector outputs; and

generate a sequence learner output indicating the detected temporal sequence.

16. The computer-readable storage medium of claim 15, wherein the sequence learner output comprises numbers representing likelihood that the input signal is caused by objects or states of the objects.

17. The computer-readable storage medium of claim 15, wherein the temporal relationships between the series of coincidence detector outputs are stored in a list.

18. The computer-readable storage medium of claim 15, wherein the coincidence detector and the sequence learner are implemented as software.

19. The computer-readable storage medium of claim 15, further comprising instructions to remove a first sequence of spatial patterns responsive to receiving a second sequence of spatial patterns that is similar to the first sequence of spatial patterns at the sequence learner in a learning stage.