Butterfly transform layer

1. A method, comprising:

receiving an image comprising a plurality of channels;

executing a convolutional neural network model configured to generate a classification of the image, wherein executing the convolutional neural network model comprises:

executing convolution on each channel of the image by a depth-wise convolutional layer to generate first data comprising a plurality of channels; and

executing a linear transformation of the first data by a butterfly transform layer comprising a plurality of sub-layers to combine the plurality of channels of the first data to generate second data, wherein the classification is based on the generated second data; and

providing the classification of the image for further processing on a computing device.

2. The method of claim 1 , wherein the plurality of sub-layers are configured to recursively calculate the second data.

3. The method of claim 1 , wherein a number of input nodes and a number of output nodes for each of the plurality of sub-layers are equal.

4. The method of claim 3 , wherein the butterfly transform layer comprises a path between each input node of a first sub-layer of the plurality of sub-layers that receives the first data and each output node of a second sub-layer of the plurality of sub-layers that outputs the second data.

5. The method of claim 1 , wherein the butterfly transform layer is further configured to execute a non-linearity function on the generated second data.

6. The method of claim 1 , wherein the butterfly transform layer is further configured to execute a batch normalization of the generated second data.

7. The method of claim 1 , wherein all nodes within each sub-layer of the plurality of sub-layers have a same out degree.

8. The method of claim 1 wherein the plurality of sub-layers comprises log(n) sub-layers, wherein n is a number of channels of the image.

9. A system, comprising:

a processor; and

a memory device containing instructions which, when executed by the processor, cause the processor to:

receive an image comprising a plurality of channels;

execute a convolutional neural network model configured to generate a classification of the image, wherein executing the convolutional neural network model comprises:

executing convolution on each channel of the image by a depth-wise convolutional layer to generate first data comprising a plurality of channels; and

executing a linear transformation of the first data by a butterfly transform layer comprising a plurality of sub-layers to combine the plurality of channels of the first data to generate second data, wherein the classification is based on the generated second data;

execute a non-linearity function on the generated second data to generate the classification of the image; and

provide the a classification of the image for further processing on a computing device.

10. The system of claim 9 , wherein the plurality of sub-layers are configured to recursively calculate the second data.

11. The system of claim 9 , wherein a number of input nodes and a number of output nodes for each of the plurality of sub-layers are equal.

12. The system of claim 9 , wherein the butterfly transform layer comprises a path between each input node of a first sub-layer of the plurality of sub-layers that receives the first data and each output node of a second sub-layer of the plurality of sub-layers that outputs the second data.

13. The system of claim 9 , wherein the butterfly transform layer is further configured to execute a batch normalization of the generated second data.

14. The system of claim 9 , wherein the plurality of sub-layers comprises log(n) sub-layers, wherein n is a number of channels of the image.

15. A computer program product comprising code stored in a non-transitory computer-readable storage medium, the code comprising:

code for receiving an image comprising a plurality of channels;

code for executing a convolutional neural network model configured to generate a classification of the image, wherein the code for executing the convolutional neural network model comprises:

code for executing convolution on each channel of the image by a depth-wise convolutional layer to generate first data comprising a plurality of channels; and

code for executing a linear transformation of the first data by a butterfly transform layer comprising a plurality of sub-layers to combine the plurality of channels of the first data to generate second data, wherein the butterfly transform layer comprises a path from each input node of the butterfly transform layer to each output node of the butterfly transform layer, and wherein the classification is based on the generated second data; and

code for providing the classification of the image for further processing on a computing device.

16. The computer program product of claim 15 , wherein a number of input nodes and a number of output nodes for each of the plurality of sub-layers are equal.

17. The computer program product of claim 15 , wherein the plurality of sub-layers are configured to recursively calculate the second data.

18. The computer program product of claim 15 , wherein code for passing the first data through the butterfly transform layer further comprises code for executing a non-linearity function on the generated second data.

19. The computer program product of claim 15 , wherein code for passing the first data through the butterfly transform layer further comprises code for normalizing the generated second data.

20. The computer program product of claim 16 , wherein the plurality of sub-layers comprises log(n) sub-layers, wherein n is a number of channels of the image.