OCR is a very simple use of a more primitive version of the same technology. You create a training set, a set of expected results, then use a training algorithm on the layers of linear algebra and matrixes to adjust them until you have a program that can accurately classify input data and produce the expected output, even with examples it has never seen before, generalizing it's functionality beyond just it's training set.
Of course it's a hell of a lot easier to match small greyscale images to a set of less than 200 glyphs than it is to match it to as vague a concept as coherent English text, and of course there are a lot of tasks with intermediary complexity like subject identification and catching and classifying defects in parts without exact measurements.
The big thing that kicked off this latest boom was production of general outputs.
Now we're able to go from general to general (a description of a picture to a picture that fits the description) instead of general to specific (picture of animal to Is Cat?Y/N with confidence interval X.)
OCR is a very simple use of a more primitive version of the same technology. You create a training set, a set of expected results, then use a training algorithm on the layers of linear algebra and matrixes to adjust them until you have a program that can accurately classify input data and produce the expected output, even with examples it has never seen before, generalizing it's functionality beyond just it's training set.
Of course it's a hell of a lot easier to match small greyscale images to a set of less than 200 glyphs than it is to match it to as vague a concept as coherent English text, and of course there are a lot of tasks with intermediary complexity like subject identification and catching and classifying defects in parts without exact measurements.
The big thing that kicked off this latest boom was production of general outputs.
Now we're able to go from general to general (a description of a picture to a picture that fits the description) instead of general to specific (picture of animal to Is Cat?Y/N with confidence interval X.)