Automatic image variety simulation for improved deep learning performance

A computer-implemented method, comprising:

• accessing, by a device operatively coupled to a processor, a failure image on which a first deep learning neural network has inaccurately performed an inferencing task;
• training, by the device and on a set of dummy images, a second deep learning neural network to learn a first visual variety of the failure image, based on a loss function having a first term and a second term, wherein the first term quantifies visual content dissimilarities between the set of dummy images and outputs predicted during training by the second deep learning neural network, and wherein the second term quantifies, at a plurality of different image scales, visual variety dissimilarities between the failure image and the outputs predicted during training by the second deep learning neural network;
• and
• executing, by the device, the second deep learning neural network on each of a set of training images on which the first deep learning neural network was trained, thereby yielding a set of first converted training images that exhibit the first visual variety of the failure image.

A computer program product for facilitating automatic image variety simulation for improved deep learning performance, the computer program product comprising a non-transitory computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

• access a failure medical image on which a first deep learning neural network has generated an inaccurate inferencing task result;
• train, on a set of non-medical images, a second deep learning neural network to learn a visual variety of the failure medical image, based on a loss function having a first term and a second term, wherein the first term quantifies visual content dissimilarities between the set of non-medical images and outputs predicted during training by the second deep learning neural network, and wherein the second term quantifies, at a plurality of different image scales, visual variety dissimilarities between the failure medical image and the outputs predicted during training by the second deep learning neural network;
• and
• execute the second deep learning neural network on each of a set of training medical images on which the first deep learning neural network was trained, thereby yielding a set of converted training medical images that exhibit the visual variety of the failure medical image.