Classifying features using a neurosynaptic system

• US 10,115,054 B2
• Filed: 07/02/2014
• Issued: 10/30/2018
• Est. Priority Date: 07/02/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Appuswamy, Rathinakumar, Esser, Steven K., Modha, Dharmendra S.
• Primary Examiner(s): Nilsson, Eric
• Application Number:
  US14/322,778
• Publication Number:
  US 20160004962A1
• Time in Patent Office: 1,581 Days
• 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

Abstract

Embodiments of the invention provide a method comprising receiving a set of features extracted from input data, training a linear classifier based on the set of features extracted, and generating a first matrix using the linear classifier. The first matrix includes multiple dimensions. Each dimension includes multiple elements. Elements of a first dimension correspond to the set of features extracted. Elements of a second dimension correspond to a set of classification labels. The elements of the second dimension are arranged based on one or more synaptic weight arrangements. Each synaptic weight arrangement represents effective synaptic strengths for a classification label of the set of classification labels. The neurosynaptic core circuit is programmed with synaptic connectivity information based on the synaptic weight arrangements. The core circuit is configured to classify one or more objects of interest in the input data.

Claims

1. A method comprising:

receiving a set of features extracted from input data;

training a linear classifier based on the set of features extracted;

generating a first matrix using the linear classifier, wherein the first matrix includes a first dimension and a second dimension, wherein each dimension includes multiple elements, wherein each element of the first dimension corresponds to a feature of the set of features extracted, and wherein each element of the second dimension corresponds to a classification label of a set of different classification labels for classifying one or more objects of interest in the input data;

arranging each element of the second dimension into a corresponding synaptic weight arrangement representing effective synaptic strengths for a classification label corresponding to the element;

programming a first neurosynaptic core circuit for dimensionality reduction by programming synaptic weights of a plurality of electronic synapse devices of the first neurosynaptic core circuit based on each synaptic weight arrangement that each element of the second dimension is arranged into; and

reducing a number of features included in the first set of features utilizing the first neurosynaptic core circuit, wherein the reduced number of features are mapped to a second neurosynaptic core circuit configured to classify the one or more objects of interest in the input data;

wherein each synaptic weight of the synaptic weight arrangement represents a feature associated with a classification label and quantized to a particular value of a range of values to facilitate the dimensionality reduction and the mapping of the reduced number of features to the second neurosynaptic core circuit such that the reduced number of features all fit onto the second neurosynaptic core circuit.

2. The method of claim 1, wherein each neurosynaptic core circuit comprises one or more electronic neurons, one or more electronic axons, and a synaptic array including a plurality of electronic synapse devices for interconnecting the neurons with the axons.

3. The method of claim 2, wherein each synaptic weight arrangement that each element of the second dimension is arranged into comprises a weight matrix including multiple integer values representing effective synaptic strengths for a classification label corresponding to the element.

4. The method of claim 3, wherein:

each axon has a corresponding axon type; and

the effective synaptic strengths include different effective synaptic strengths for different axon types.

5. The method of claim 2, wherein:

each element of the first dimension represents a row of the first matrix;

each element of the second dimension represents a column of the first matrix; and

the arranging further comprises;

dividing at least one column of the first matrix into multiple columns.

6. The method of claim 5, wherein each column of the first matrix is divided into multiple columns based on a Latin square.

7. The method of claim 6, further comprising:

mapping the first matrix to a synaptic array of the first neurosynaptic core circuit to reduce spiking of the neurons of the core circuit.

8. The method of claim 1, wherein the input data comprises one of an image sequence, an audio sequence, and a video sequence.

9. A system comprising:

at least one processor; and

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

receiving a set of features extracted from input data;

training a linear classifier based on the set of features extracted;

generating a first matrix using the linear classifier, wherein the first matrix includes a first dimension and a second dimension, wherein each dimension includes multiple elements, wherein each element of the first dimension corresponds to a feature of the set of features extracted, and wherein each element of the second dimension corresponds to a classification label of a set of different classification labels for classifying one or more objects of interest in the input data;

arranging each element of the second dimension into a corresponding synaptic weight arrangement representing effective synaptic strengths for a classification label corresponding to the element;

programming a first neurosynaptic core circuit for dimensionality reduction by programming synaptic weights of a plurality of electronic synapse devices of the first neurosynaptic core circuit based on each synaptic weight arrangement that each element of the second dimension is arranged into; and

reducing a number of features included in the first set of features utilizing the first neurosynaptic core circuit, wherein the reduced number of features are mapped to a second neurosynaptic core circuit configured to classify the one or more objects of interest in the input data;

wherein each synaptic weight of the synaptic weight arrangement represents a feature associated with a classification label and quantized to a particular value of a range of values to facilitate the dimensionality reduction and the mapping of the reduced number of features to the second neurosynaptic core circuit such that the reduced number of features all fit onto the second neurosynaptic core circuit.

10. The system of claim 9, wherein each neurosynaptic core circuit comprises one or more electronic neurons, one or more electronic axons, and a synaptic array including a plurality of electronic synapse devices for interconnecting the neurons with the axons.

11. The system of claim 10, wherein each synaptic weight arrangement that each element of the second dimension is arranged into comprises a weight matrix including multiple integer values representing effective synaptic strengths for a classification label corresponding to the element.

12. The system of claim 11, wherein:

each axon has a corresponding axon type; and

the effective synaptic strengths include different effective synaptic strengths for different axon types.

13. The system of claim 10, wherein:

each element of the first dimension represents a row of the first matrix;

each element of the second dimension represents a column of the first matrix; and

the arranging further comprises;

dividing at least one column of the first matrix into multiple columns.

14. The system of claim 13, wherein a column of the first matrix is divided into multiple columns based on a Latin square.

15. The system of claim 14, wherein the first matrix is mapped to a synaptic array of the first neurosynaptic core circuit to reduce spiking of the neurons of the core circuit.

16. The system of claim 9, wherein the input data comprises one of an image sequence, an audio sequence, and a video sequence.

17. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor of a machine, cause the machine to perform operations comprising:

receiving a set of features extracted from input data;

training a linear classifier based on the set of features extracted;

generating a first matrix using the linear classifier, wherein the first matrix includes a first dimension and a second dimension, wherein each dimension includes multiple elements, wherein each element of the first dimension corresponds to a feature of the set of features extracted, and wherein each element of the second dimension corresponds to a classification label of a set of different classification labels for classifying one or more objects of interest in the input data;

arranging each element of the second dimension into a corresponding synaptic weight arrangement representing effective synaptic strengths for a classification label corresponding to the element;

programming a first neurosynaptic core circuit for dimensionality reduction by programming synaptic weights of a plurality of electronic synapse devices of the first neurosynaptic core circuit based on each synaptic weight arrangement that each element of the second dimension is arranged into; and

reducing a number of features included in the first set of features utilizing the first neurosynaptic core circuit, wherein the reduced number of features are mapped to a second neurosynaptic core circuit configured to classify the one or more objects of interest in the input data;

wherein each synaptic weight of the synaptic weight arrangement represents a feature associated with a classification label and quantized to a particular value of a range of values to facilitate the dimensionality reduction and the mapping of the reduced number of features to the second neurosynaptic core circuit such that the reduced number of features all fit onto the second neurosynaptic core circuit.