Training a neural network adapter

1. A system comprising:

a processor to:

detect a trained neural network application;

detect a set of images from a medical device, wherein the trained neural network application is not trained with the set of images;

train an adapter network based on the trained neural network application and the set of images, wherein the adapter network is to be trained by freezing weights of the trained neural network application and modifying weights of the adapter network, wherein the processor is to initialize the weights of the adapter network to weights of a pass through function; and

use the trained adapter network to process at least one additional image, the at least one processed additional image to be transmitted to the trained neural network application to generate an output signal.

2. The system of claim 1 , wherein the output signal indicates a medical condition corresponding to the at least one additional image, wherein the at least one image is an x-ray, an ultrasound image, or an MRI image.

3. The system of claim 1 , wherein the processor is to use the trained neural network application as a loss function and use a backpropagation technique to determine the weights of the adapter network.

4. The system of claim 1 , wherein the processed additional image comprises a modification to the at least one additional image, the modification comprising a modified color, a modified hue, a modified saturation value, or a modified brightness.

5. The system of claim 1 , wherein the at least one processed additional image comprises a modified scaling or a modified skew.

6. The system of claim 1 , wherein the at least one processed additional image comprises characteristics shared with a sample image from a first image capturing device, wherein the at least one processed additional image is detected from a second image capturing device.

7. The system of claim 1 , wherein the processor is to generate the at least one processed additional image by applying a modified sigmoid function associated with the adapter network to the at least one additional image.

8. The system of claim 7 , wherein the modified sigmoid function corresponds to a degradation of the medical device over time, and the at least one processed additional image represents characteristics shared with output from the medical device at an earlier date.

9. The system of claim 1 , wherein the adapter network comprises a plurality of rectified linear units and a plurality of convolutional neural networks.

10. A method comprising:

detecting a trained neural network application;

detecting a set of images from a medical device, wherein the trained neural network application is not trained with the set of images;

training an adapter network based on the trained neural network application and the set of images, wherein the adapter network is to be trained by freezing weights of the trained neural network application and modifying weights of the adapter network, wherein the weights of the adapter network are initialized to weights of a pass through function; and

using the trained adapter network to process at least one additional image, the at least one processed additional image to be transmitted to the trained neural network application to generate an output signal.

11. The method of claim 10 , wherein the output signal indicates a medical condition corresponding to the at least one additional image, wherein the at least one additional image is an x-ray, an ultrasound image, or an MRI image.

12. The method of claim 10 , comprising using the trained neural network application as a loss function and use a backpropagation technique to determine the weights of the adapter network.

13. The method of claim 10 , wherein the at least one processed additional image comprises a modification to the at least one image, the modification comprising a modified color, a modified hue, a modified saturation value, or a modified brightness.

14. The method of claim 10 , wherein the at least one processed additional image comprises a modified scaling or a modified skew.

15. The method of claim 10 , wherein the at least one processed additional image comprises characteristics shared with a sample image from a first image capturing device, wherein the at least one processed additional image is detected from a second image capturing device.

16. The method of claim 10 , comprising generating the at least one processed additional image by applying a modified sigmoid function associated with the adapter network to the at least one additional image.

17. A computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, wherein the program instructions are executable by a processor to cause the processor to:

detect a trained neural network application;

detect a set of images from a medical device, wherein the trained neural network application is not trained with the set of images;

train an adapter network based on the trained neural network application and the set of images, wherein the adapter network is to be trained by freezing weights of the trained neural network application and modifying weights of the adapter network, wherein the processor is to initialize the weights of the adapter network to weights of a pass through function; and

use the trained adapter network to process at least one additional image, the processed additional image to be transmitted to the trained neural network application to generate an output signal.

18. The computer program product of claim 17 , wherein the output signal indicates a medical condition corresponding to the at least one additional image, wherein the at least one additional image is an x-ray, an ultrasound image, or an MRI image.