Method and system for distributed coding and learning in neuromorphic networks for pattern recognition

1. A system for pattern recognition applied to neuromorphic hardware, the system comprising:

one or more processors and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the one or more processors perform an operation of:

training a spiking neural network that is applied to neuromorphic hardware, wherein training the spiking neural network comprises operations of:

generating, with each excitatory neuron in an excitatory layer of the spiking neural network, a spike train of neuron spikes for a plurality of training patterns, each training pattern belonging to a pattern class;

generating a spiking rate distribution of excitatory neurons for each pattern class;

normalizing each spiking rate distribution of excitatory neurons with a L2 norm; and

generating a class template for each pattern class from the normalized spiking rate distributions.

2. The system as set forth in claim 1 , where in generating the spiking rate distribution of excitatory neurons for each pattern class, the neuron spikes from all training patterns in a pattern class are summed.

3. The system as set forth in claim 1 , wherein when generating the spike trains, multiple excitatory neurons can fire at the same time to allow for distributed coding.

4. The system as set forth in claim 1 , where in generating the spike train of neuron spikes for the plurality of training patterns, each input neuron in an input layer of the spiking neural network generates a spike train according to an intensity value for a corresponding feature channel, wherein the greater the intensity value, the more neuron spikes are generated.

5. A system for pattern recognition applied to neuromorphic hardware, the system comprising:

one or more processors and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the one or more processors perform operations of:

using a trained spiking neural network applied to neuromorphic hardware to classify an unlabeled input pattern, wherein classifying the unlabeled input pattern comprises operations of:

using L2 normalization, classifying an unlabeled input pattern using class templates generated for a plurality of pattern classes from normalized spike rate distributions of excitatory neurons in the spiking neural network; and

controlling a mechanical component of an autonomous device based on classification of the unlabeled input pattern.

6. The system as set forth in claim 5 , where in classifying the unlabeled input pattern, the one or more processors further perform operations of:

comparing an excitatory neuron spiking rate pattern of the unlabeled input pattern with each class template; and

determining a match between the excitatory neuron spiking rate pattern of the unlabeled input pattern and a class template.

7. The system as set forth in claim 6 , where in classifying the unlabeled input pattern using the class templates, the one or more processors further perform an operation of predicting a class label for the unlabeled input pattern from its matched class template.

8. A computer implemented method for pattern recognition applied to neuromorphic hardware, the method comprising an act of:

causing one or more processors to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of:

training a spiking neural network that is applied to neuromorphic hardware, wherein training the spiking neural network comprises operations of:

generating, with each excitatory neuron in an excitatory layer of the spiking neural network, a spike train of neuron spikes for a plurality of training patterns, each training pattern belonging to a pattern class;

generating a spiking rate distribution of excitatory neurons for each pattern class;

normalizing each spiking rate distribution of excitatory neurons with a L2 norm; and

generating a class template for each pattern class from the normalized spiking rate distributions.

9. The method as set forth in claim 8 , where in generating the spiking rate distribution of excitatory neurons for each pattern class, the neuron spikes from all training patterns in a pattern class are summed.

10. The method as set forth in claim 8 , wherein when generating the spike trains, multiple excitatory neurons can fire at the same time.

11. The method as set forth in claim 8 , where in generating the spike train of neuron spikes for the plurality of training patterns, each input neuron in an input layer of the spiking neural network generates a spike train according to an intensity value for a corresponding feature channel, wherein the greater the intensity value, the more neuron spikes are generated.

12. A computer implemented method for pattern recognition applied to neuromorphic hardware, the method comprising an act of:

causing one or more processors to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of:

using a trained spiking neural network applied to neuromorphic hardware to classify an unlabeled input pattern, wherein classifying the unlabeled input pattern comprises operations of:

using L2 normalization, classifying an unlabeled input pattern using class templates generated for a plurality of pattern classes from normalized spike rate distributions of excitatory neurons in the spiking neural network; and

controlling a mechanical component of an autonomous device based on classification of the unlabeled input pattern.

13. The method as set forth in claim 12 , where in classifying the unlabeled input pattern, the one or more processors further perform operations of:

comparing an excitatory neuron spiking rate pattern of the unlabeled input pattern with each class template; and

determining a match between the excitatory neuron spiking rate pattern of the unlabeled input pattern and a class template.

14. The method as set forth in claim 13 , where in classifying the unlabeled input pattern using the class templates, the one or more processors further perform an operation of predicting a class label for the unlabeled input pattern from its matched class template.

15. A computer program product for pattern recognition applied to neuromorphic hardware, the computer program product comprising:

computer-readable instructions stored on a non-transitory computer-readable medium that are executable by a computer having one or more processors for causing the processor to perform operations of:

training a spiking neural network that is applied to neuromorphic hardware, wherein training the spiking neural network comprises operations of:

generating, with each excitatory neuron in an excitatory layer of the spiking neural network, a spike train of neuron spikes for a plurality of training patterns, each training pattern belonging to a pattern class;

generating a spiking rate distribution of excitatory neurons for each pattern class;

normalizing each spiking rate distribution of excitatory neurons with a L2 norm; and

generating a class template for each pattern class from the normalized spiking rate distributions.

16. The computer program product as set forth in claim 15 , where in generating the spiking rate distribution of excitatory neurons for each pattern class, the neuron spikes from all training patterns in a pattern class are summed.

17. The computer program product as set forth in claim 15 , wherein when generating the spike trains, multiple excitatory neurons can fire at the same time.

18. The computer program product as set forth in claim 15 , where in generating the spike train of neuron spikes for the plurality of training patterns, each input neuron in an input layer of the spiking neural network generates a spike train according to an intensity value for a corresponding feature channel, wherein the greater the intensity value, the more neuron spikes are generated.

19. A computer program product for pattern recognition applied to neuromorphic hardware, the computer program product comprising:

computer-readable instructions stored on a non-transitory computer-readable medium that are executable by a computer having one or more processors for causing the processor to perform operations of:

using a trained spiking neural network applied to neuromorphic hardware to classify an unlabeled input pattern, wherein classifying the unlabeled input pattern comprises operations of:

using L2 normalization, classifying an unlabeled input pattern using class templates generated for a plurality of pattern classes from normalized spike rate distributions of excitatory neurons in the spiking neural network; and

controlling a mechanical component of an autonomous device based on classification of the unlabeled input pattern.

20. The computer program product as set forth in claim 19 , where in classifying the unlabeled input pattern, the one or more processors further perform operations of:

comparing an excitatory neuron spiking rate pattern of the unlabeled input pattern with each class template; and

determining a match between the excitatory neuron spiking rate pattern of the unlabeled input pattern and a class template.

21. The computer program product as set forth in claim 20 , where in classifying the unlabeled input pattern using the class templates, the one or more processors further perform an operation of predicting a class label for the unlabeled input pattern from its matched class template.