UNSUPERVISED, SUPERVISED AND REINFORCED LEARNING VIA SPIKING COMPUTATION

• US 20170017878A1
• Filed: 11/18/2015
• Published: 01/19/2017
• 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.
• Granted Patent: US 10,019,669 B2
• Time in Patent Office: Days
• 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:

receiving an input at a neural network comprising a plurality of neural modules interconnected via a plurality of weighted synaptic connections;

extracting one or more input features from the input as the input propagates through the neural network via at least one of the weighted synaptic connections;

associating the one or more input features with the input; and

in response to receiving a version of the input at the neural network, producing an approximation of the version of the input by reproducing the input.

2. The method of claim 1, wherein:

each neural module comprises multiple neurons; and

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

3. The method of claim 1, wherein the version of the input is a noisy version.

4. The method of claim 2, further comprising:

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

5. The method of claim 1, further comprising:

applying at least one learning rule to the at least one weighted synaptic connection, such that the input is amplified as the input propagates through the neural network via the at least one weighted synaptic connection.

6. A system comprising a computer processor, a computer-readable hardware storage medium, and program code embodied with the computer-readable hardware storage medium for execution by the computer processor to implement a method comprising:

receiving an input at a neural network comprising a plurality of neural modules interconnected via a plurality of weighted synaptic connections;

extracting one or more input features from the input as the input propagates through the neural network via at least one of the weighted synaptic connections;

applying at least one learning rule to the at least one weighted synaptic connection, such that the input is amplified as the input propagates through the neural network via the at least one weighted synaptic connection;

associating the one or more input features with the input; and

in response to receiving a version of the input at the neural network, producing an approximation of the version of the input by reproducing the input.

7. The system of claim 6, wherein:

each neural module comprises multiple neurons; and

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

8. The system of claim 6, wherein the version of the input is a noisy version.

9. The system of claim 7, the method further comprising:

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

10. The system of claim 6, the method further comprising:

applying at least one learning rule to the at least one weighted synaptic connection, such that the input is amplified as the input propagates through the neural network via the at least one weighted synaptic connection.

11. A computer program product comprising a computer-readable hardware storage device having program code embodied therewith, the program code being executable by a computer to implement a method comprising:

receiving an input at a neural network comprising a plurality of neural modules interconnected via a plurality of weighted synaptic connections;

extracting one or more input features from the input as the input propagates through the neural network via at least one of the weighted synaptic connections;

applying at least one learning rule to the at least one weighted synaptic connection, such that the input is amplified as the input propagates through the neural network via the at least one weighted synaptic connection;

associating the one or more input features with the input; and

in response to receiving a version of the input at the neural network, producing an approximation of the version of the input by reproducing the input.

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

each neural module comprises multiple neurons; and

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

13. The computer program product of claim 11, wherein the version of the input is a noisy version.

14. The computer program product of claim 12, the method further comprising:

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

15. The computer program product of claim 11, the method further comprising:

applying at least one learning rule to the at least one weighted synaptic connection, such that the input is amplified as the input propagates through the neural network via the at least one weighted synaptic connection.