Methods, architecture, and apparatus for implementing machine intelligence and hierarchical memory systems

• US 9,530,091 B2
• Filed: 04/03/2012
• Issued: 12/27/2016
• Est. Priority Date: 12/10/2004
• Status: Active Grant

• Current Assignee: Numenta, Inc.
• Original Assignee: Numenta, Inc.
• Inventors: Hawkins, Jeffrey, George, Dileep
• Primary Examiner(s): Wong, Lut
• Application Number:
  US13/438,670
• Publication Number:
  US 20120197823A1
• Time in Patent Office: 1,729 Days
• Field of Search: 706/12
• US Class Current: 1/1
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Sophisticated memory systems and intelligent machines may be constructed by creating an active memory system with a hierarchical architecture. Specifically, a system may comprise a plurality of individual cortical processing units arranged into a hierarchical structure. Each individual cortical processing unit receives a sequence of patterns as input. Each cortical processing unit processes the received input sequence of patterns using a memory containing previously encountered sequences with structure and outputs another pattern. As several input sequences are processed by a cortical processing unit, it will therefore generate a sequence of patterns on its output. The sequence of patterns on its output may be passed as an input to one or more cortical processing units in next higher layer of the hierarchy. A lowest layer of cortical processing units may receive sensory input from the outside world. The sensory input also comprises a sequence of patterns.

Claims

1. A computer system comprising:

a processor; and

a memory coupled to the processor, the memory comprising;

a first lower level processing unit configured to receive a first portion of an input data at a first time and generate a first output representing information about temporal sequences of spatial patterns in the first portion of the input data corresponding to first learned temporal sequences of spatial patterns;

a second lower level processing unit configured to receive a second portion of the input data at the first time and generate a second output representing information about temporal sequences of spatial patterns in the second portion of the input data corresponding to second learned temporal sequences of spatial patterns; and

an upper level processing unit associated with the first and second lower level processing units, the upper level processing unit configured to generate a third output based on the first and the second outputs, the third output sent to at least one of the first and second lower level processing unit as a first prediction of spatial patterns in the first portion of the input data to be received by the first lower level processing unit or the second portion of the input data to be received by the second lower level processing node at a second time subsequent to the first time.

2. The computer system of claim 1, further comprising one or more sensory units configured to provide the input data to the first lower level processing unit and the second lower level processing unit.

3. The computer system of claim 1, wherein the first lower level processing unit, the second lower level processing unit, and the upper level processing unit each comprises a probability table for generating the first output, the second output or the third output.

4. The computer system of claim 1, wherein the first lower level processing unit, the second lower level processing unit, and the upper level processing unit form a Bayesian belief net.

5. The computer system of claim 1, wherein the first lower level processing unit is configured to generate a second prediction output based on the first portion of the input data received at a first time, the second prediction output representing predicted input data to be received at the second time, the second prediction output fed back to the first lower level processing unit at the second time to generate the first output at a third time subsequent to the second time.

6. The computer system of claim 5, wherein the prediction output resolves ambiguities in the first portion of the input data received at the first lower level processing unit.

7. The computer system of claim 1, wherein the first learned temporal sequences are temporal sequences in a training input data detected more frequently by the first lower level processing unit than other temporal sequences in the training input data and the second learned temporal sequences are temporal sequences in the training input detected more frequently by the second lower level processing unit than other temporal sequences in the training input data.

8. The computer system of claim 7, wherein the training input data comprises animated movies.

9. The computer system of claim 1, wherein the upper level processing unit is configured to learn temporal sequences in a concatenation of the first output and the second output.

10. The computer system of claim 1, wherein the upper level processing unit determines whether sequences of combinations of the first output and the second output correspond to third learned sequences of patterns, the first and second learned sequences of patterns representing a first level of structures in the input data, and the third learned sequences of patterns representing a second level of structures in the input data that are formed by the first level of structures.

11. The computer system of claim 10, wherein the first level of structures comprise vertical lines, horizontal lines, corners, boxes, and parallel lines;

and the second level of structures comprise letters.

12. The computer system of claim 1, wherein the first learned temporal sequences are temporal sequences in a training input data detected more frequently by the first lower level processing unit than other temporal sequences in the training input data and the second learned temporal sequences are temporal sequences in the training input detected more frequently by the second lower level processing unit than other temporal sequences in the training input data.

13. A computer-implemented method comprising:

at a first lower level processing unit, generating a first output representing information about temporal sequences of spatial patterns in a first portion of an input data received at a first time corresponding to first learned temporal sequences of spatial patterns;

at a second lower level processing unit, generating a second output representing information about temporal sequences of spatial patterns in a second portion of the input data received at the first time corresponding to second learned temporal sequences of spatial patterns;

at an upper level processing unit, generating a third output based on the first and the second outputs, the third output representing information about causes to the input data; and

sending the third output to at least one of the first and second lower level processing unit as a first prediction of spatial patterns in the first portion of the input data to be received by the first lower level processing unit or the second portion of the input data to be received by the second lower level processing node at a second time subsequent to the first time.

14. The method of claim 13, wherein the input data to the first lower level processing unit and the second lower level processing unit is generated by one or more sensor units.

15. The method of claim 13, wherein the first lower level processing unit, the second lower level processing unit, and the upper level processing unit form a Bayesian belief net.

16. The method of claim 13, further comprising:

generating a second prediction output at the first lower level processing unit based on the first portion of the input data received at the first time, the prediction output representing predicted input data to be received at the second time; and

feeding the second prediction output back to the first lower level processing unit at the second time to generate the first output at a third time subsequent to the second time.

17. The method of claim 16, further comprising resolving ambiguities in the first portion of the input data received at the first lower level processing unit using the prediction output.

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

generate, at a first lower level processing unit, a first output representing information about temporal sequences of spatial patterns in a first portion of an input data received at a first time corresponding to first learned temporal sequences of spatial patterns;

generate, at a second lower level processing unit, a second output representing information about temporal sequences of spatial patterns in a second portion of the input data received at the first time corresponding to second learned temporal sequences of spatial patterns;

generate, at an upper level processing unit, a third output based on the first and the second outputs, the third output representing information about causes to the input data; and

send the third output to at least one of the first and second lower level processing unit as a first prediction of spatial patterns in the first portion of the input data to be received by the first lower level processing unit or the second portion of the input data to be received by the second lower level processing node at a second time subsequent to the first time.