A large number of business problems relating to predictions can be transformed into or directly represented as computer vision and image pattern recognition problems. A vast array of deep learning technology can then be applied to get highly accurate models.
Described below are insights derived while we implemented a specific business problem to classify a part image into a good part or a bad part.
A two dimensional image scan of the part would have a stain like splotch that would appear with certain shape in the image. We only had access to a few hundred images and so the deep learning model was built without access to the millions of images on which standard datasets like ImageNet and CIFAR 10 are built.