Unsupervised, supervised, and reinforced learning via spiking computation

• US 9,390,372 B2
• Filed: 09/23/2014
• Issued: 07/12/2016
• Est. Priority Date: 09/16/2011
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Modha, Dharmendra S.
• Primary Examiner(s): CHANG, LI WU
• Application Number:
  US14/494,372
• Publication Number:
  US 20150363690A1
• Time in Patent Office: 658 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/088 Non-supervised learning, e.g. competitive learning

Abstract

The present invention relates to unsupervised, supervised and reinforced learning via spiking computation. The neural network comprises a plurality of neural modules. Each neural module comprises multiple digital neurons such that each neuron in a neural module has a corresponding neuron in another neural module. An interconnection network comprising a plurality of edges interconnects the plurality of neural modules. Each edge interconnects a first neural module to a second neural module, and each edge comprises a weighted synaptic connection between every neuron in the first neural module and a corresponding neuron in the second neural module.

Claims

1. A method for feature extraction, comprising:

providing sensory input to a first neural module in a neural network comprising a plurality of neural modules interconnected via a plurality of weighted synaptic connections;

extracting one or more input features from said sensory input as said sensory input propagates from said first neural module through said neural network in a first direction via at least one of said plurality of weighted synaptic connections;

providing motor output to a second neural module in said neural network;

extracting one or more output features from said motor output as said motor output propagates from said second neural module through said neural network in a second direction via at least one of said plurality of weighted synaptic connections, wherein said first direction is opposite of said second direction; and

associating said one or more input features with said one or more output features.

2. The method of claim 1, wherein:

each neural module comprises multiple digital neurons; and

each neuron in each neural module is connected to a corresponding neuron in another neural module via a weighted synaptic connection.

3. The method of claim 1, further comprising:

applying a learning rule to at least one of said plurality of weighted synaptic connections, such that said sensory input is amplified as said sensory input propagates from said first neural module through said neural network via said at least one weighted synaptic connection.

4. The method of claim 1, further comprising:

applying a learning rule to at least one of said plurality of weighted synaptic connections, such that said motor output is amplified as said motor output propagates from said second neural module through said neural network via said at least one weighted synaptic connection.

5. The method of claim 2, further comprising:

adaptively updating weights of said plurality of weighted synaptic connections as a function of firing events of interconnected neurons.

6. A system, comprising:

at least one processor; and

a non-transitory processor-readable memory device storing instructions that when executed by said at least one processor causes said at least one processor to perform operations including;

receiving, at a first neural module of a plurality of neural modules of a neural network, sensory input;

receiving, at a second neural module of said plurality of neural modules, motor output;

extracting one or more input features from said sensory input as said sensory input propagates from said first neural module through said neural network in a first direction via at least one of a plurality of weighted synaptic connections interconnecting said plurality of neural modules;

extracting one or more output features from said motor output as said motor output propagates from said second neural module through said neural network in a second direction via at least one of said plurality of weighted synaptic connections, wherein said first direction is opposite of said second direction; and

associating said one or more input features with said one or more output features.

7. The system of claim 6, wherein:

each neural module comprises multiple digital neurons; and

each neuron in each neural module is connected to a corresponding neuron in another neural module via a weighted synaptic connection of said plurality of weighted synaptic connections.

8. The system of claim 6, wherein:

a learning rule is applied to at least one of said plurality of weighted synaptic connections, such that said sensory input is amplified as said sensory input propagates from said first neural module through said neural network via said at least one weighted synaptic connection.

9. The system of claim 6, wherein:

a learning rule is applied to at least one of said plurality of weighted synaptic connections, such that said motor output is amplified as said motor output propagates from said second neural module through said neural network via said at least one weighted synaptic connection.

10. The system of claim 7, wherein:

weights of said plurality of weighted synaptic connections are adaptively updated as a function of firing events of interconnected neurons.

11. A non-transitory computer program product on a computer-readable medium for feature extraction, the computer program product comprising instructions which when executed on a computer cause the computer to perform operations including:

providing sensory input to a first neural module in a neural network comprising a plurality of neural modules interconnected via a plurality of weighted synaptic connections;

extracting one or more input features from said sensory input as said sensory input propagates from said first neural module through said neural network in a first direction via at least one of said plurality of weighted synaptic connections;

providing motor output to a second neural module in said neural network;

extracting one or more output features from said motor output as said motor output propagates from said second neural module through said neural network in a second direction via at least one of said plurality of weighted synaptic connections, wherein said first direction is opposite of said second direction; and

associating said one or more input features with said one or more output features.

12. The computer program product of claim 11, wherein:

each neural module comprises multiple digital neurons; and

each neuron in each neural module is connected to a corresponding neuron in another neural module via a weighted synaptic connection.

13. The computer program product of claim 11, wherein the computer program product further comprises instructions which when executed on a computer cause the computer to perform operations including:

applying a learning rule to at least one of said plurality of weighted synaptic connections, such that said sensory input is amplified as said sensory input propagates from said first neural module through said neural network via said at least one weighted synaptic connection.

14. The computer program product of claim 11, wherein the computer program product further comprises instructions which when executed on a computer cause the computer to perform operations including:

applying a learning rule to at least one of said plurality of weighted synaptic connections, such that said motor output is amplified as said motor output propagates from said second neural module through said neural network via said at least one weighted synaptic connection.