Michael Konstantinov
21 Sep 2019
Data Science and Machine Learning

I’m Not a Robot! AI VS CAPTCHA: Life After Turing Test

I’ll tell you about interesting approaches to the automatic CAPTCHA solution (Computers and Humans Apart), and also about the history and future of this test. We will consider not only the solutions of localization and segmentation problems, but also the construction of a neural machine translator that maps noisy and incomplete images into readable text. We will look at how to integrate the various neural networks used in image processing and neural machine translation to achieve high recognition accuracy to automatically solve captchas at the human level. We consider the current models and their work on the example of modern CAPTCHA from Google and Yandex. I am interested in this because it combines various models of machine learning so that the data are transformed and information can flow like a tensor in a computational graph without intermediate hard decisions or classifications.