Transform for a neurosynaptic core circuit

1. A method for feature extraction, comprising:

generating synaptic connectivity information for a neurosynaptic core circuit, wherein the core circuit comprises one or more electronic neurons, one or more electronic axons, and an interconnect fabric including a plurality of synapse devices for interconnecting said one or more electronic neurons with said one or more electronic axons;

initializing said interconnect fabric based on the synaptic connectivity information generated;

receiving input via said one or more electronic axons; and

extracting a set of features from the input received, wherein the set of features extracted comprises a set of features with reduced correlation.

2. The method of claim 1 , wherein:

the synaptic connectivity information for the neurosynaptic core circuit is based on a Kronecker product between a first matrix and a second matrix.

3. The method of claim 2 , wherein:

the first matrix is an orthogonal matrix.

4. The method of claim 2 , wherein:

the second matrix is a random binary matrix comprising entries that are independent and binomially distributed.

5. The method of claim 2 , wherein:

the second matrix is learned based on a distribution of the input received using one or more machine learning techniques.

6. The method of claim 2 , wherein:

the synaptic connectivity information generated comprises:

synaptic weights for said plurality of synapse devices; and

effective synaptic strengths for said one or more electronic axons; and

each feature represents a distinct characteristic of the input.

7. The method of claim 6 , further comprising:

for each electronic neuron:

adjusting a membrane potential variable of said electronic neuron when said electronic neuron receives input from an electronic axon, wherein said membrane potential variable is adjusted based on an effective synaptic strength for said electronic axon and a synaptic weight of a synapse device interconnecting said electronic neuron with said electronic axon.

8. A method for reducing correlation between features, comprising:

generating synaptic connectivity information for a neurosynaptic core circuit,

wherein the core circuit comprises one or more electronic neurons, one or more electronic axons, and an interconnect fabric including a plurality of synapse devices for interconnecting said one or more electronic neurons with said one or more electronic axons;

initializing said interconnect fabric based on the synaptic connectivity information generated;

receiving multiple input features via said one or more electronic axons; and

generating a set of features with reduced correlation for said multiple input features by combining said multiple input features based on the synaptic connectivity information generated.

9. The method of claim 8 , wherein:

the synaptic connectivity information for the neurosynaptic core circuit is based on a Kronecker product between a first matrix and a second matrix.

10. The method of claim 9 , wherein:

the first matrix is an orthogonal matrix.

11. The method of claim 9 , wherein:

the second matrix is a random binary matrix comprising entries that are independent and binomially distributed.

12. The method of claim 9 , wherein:

the second matrix is learned based on a distribution of said multiple input features using one or more machine learning techniques.

13. The method of claim 9 , wherein:

the synaptic connectivity information generated comprises:

synaptic weights for said plurality of synapse devices; and

effective synaptic strengths for said one or more electronic axons; and

each feature represents a distinct characteristic of the said multiple input features.

14. The method of claim 13 , further comprising:

for each electronic neuron:

adjusting a membrane potential variable of said electronic neuron when said electronic neuron receives an input feature from an electronic axon, wherein said membrane potential variable is adjusted based on an effective synaptic strength for said electronic axon and a synaptic weight of a synapse device interconnecting said electronic neuron with said electronic axon.

15. A neurosynaptic core circuit for feature extraction, comprising:

one or more electronic neurons;

one or more electronic axons for receiving input; and

an interconnect fabric including a plurality of synapse devices for interconnecting said one or more electronic neurons with said one or more electronic axons;

wherein said interconnect fabric is initialized with synaptic connectivity information; and

wherein said one or more electronic neurons generate output comprising a set of features with reduced correlation extracted from the input received based on the synaptic connectivity information.

16. The neurosynaptic core circuit of claim 15 , wherein:

the synaptic connectivity information for the neurosynaptic core circuit is based on a Kronecker product between a first matrix and a second matrix;

the synaptic connectivity information generated comprises synaptic weights for said plurality of synapse devices and effective synaptic strengths for said one or more electronic axons; and

for each electronic neuron, a membrane potential variable of said electronic neuron is adjusted when said electronic neuron receives input from an electronic axon, wherein said membrane potential variable is adjusted based on an effective synaptic strength for said electronic axon and a synaptic weight of a synapse device interconnecting said electronic neuron with said electronic axon.

17. The neurosynaptic core circuit of claim 15 , wherein:

each feature represents a distinct characteristic of the input.

18. A neurosynaptic core circuit for reducing correlation between features, comprising:

one or more electronic neurons;

one or more electronic axons for receiving multiple input features; and

an interconnect fabric including a plurality of synapse devices for interconnecting said one or more electronic neurons with said one or more electronic axons;

wherein said interconnect fabric is initialized with synaptic connectivity information; and

wherein said one or more electronic neurons generate a set of features with reduced correlation for said multiple input features by combining said multiple input features based on the synaptic connectivity information.

19. The neurosynaptic core circuit of claim 18 , wherein:

the synaptic connectivity information for the neurosynaptic core circuit is based on a Kronecker product between a first matrix and a second matrix;

the synaptic connectivity information generated comprises synaptic weights for said plurality of synapse devices and effective synaptic strengths for said one or more electronic axons; and

for each electronic neuron, a membrane potential variable of said electronic neuron is adjusted when said electronic neuron receives input from an electronic axon, wherein said membrane potential variable is adjusted based on an effective synaptic strength for said electronic axon and a synaptic weight of a synapse device interconnecting said electronic neuron with said electronic axon.

20. The neurosynaptic core circuit of claim 18 , wherein:

each feature represents a distinct characteristic of said multiple input features.