Scheduling system and method in a hierarchical temporal memory based system

• US 7,941,392 B2
• Filed: 02/28/2007
• Issued: 05/10/2011
• Est. Priority Date: 02/28/2007
• Status: Active Grant

• Current Assignee: Numenta, Inc.
• Original Assignee: Numenta, Inc.
• Inventors: Saphir, William Cooper
• Primary Examiner(s): Holmes; Michael B.
• Assistant Examiner(s): Kennedy; Adrian L
• Application Number:
  US11/713,157
• Publication Number:
  US 20080205280A1
• Time in Patent Office: 1,532 Days
• Field of Search: None
• US Class Current: 706/52
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs

Abstract

According to one aspect of one or more embodiments of the present invention, a system comprises: an HTM network executable at least in part on multiple node processing units (NPUs). In one embodiment the NPUs include one or more nodes, each of which can be executed by its NPU. In one embodiment, the present invention includes a technique for coordinating and scheduling HTM computation across one or more CPUs which (1) enables concurrent computation (2) does not require a central point of control (e.g. a controller entity that “orchestrates” the computation), (3) does not require global synchronization, (4) in some embodiments ensures that the same results are achieved whether the nodes are executed in parallel or serially.

Claims

1. A method for performing phase based scheduling in a network having two or more node processing units each node processing unit having at least one node, with each of the at least one node having a single output buffer, the method comprising the steps of:

assigning, prior to performing a compute operation on said two or more node processing units, one or more phase values to each of said nodes, wherein said phase values define a first order such that if said nodes are computed in the first order that data dependency requirements are satisfied within the two or more node processing units and wherein two or more nodes can have an identical phase value; and

synchronizing the two or more node processing units, having the steps of;

(a) selecting a phase epoch in each of said node processing units wherein a phase epoch is a period in which all nodes having a particular phase value are computed;

(b) selecting an uncomputed node in said epoch;

(c) performing the compute operation on said selected node if said selected node is enabled;

(d) outputting a result of said compute operation;

(e) repeating steps (b) through (d) for each uncomputed node in said epoch; and

(f) repeating steps (a) through (e) for phase epoch.

2. The method of claim 1, wherein two or more phase values are assigned to at least one node.

3. The method of claim 1, wherein the synchronizing step further comprises the steps of:

(a) sending a READY message to all upstream node processing units, prior to performing said compute operation;

(b) waiting for said READY message from all downstream node processing units;

(c) sending a DONE message to all downstream node processing units after performing said compute operation of all nodes in said selected phased epoch; and

(d) waiting for said DONE message from all upstream node processing units.

4. The method of claim 1, further comprising the steps of the node processing units automatically mirroring the result, by making the result accessible to a downstream node processing unit which is one of the node processing units.

5. The method of claim 1, wherein the synchronization step does not send a synchronization message between said two or more node processing units.