As machine learning algorithms continue to advance, the need for good, accurately annotated datasets is becoming increasingly apparent. With less and less room for optimization of the models themselves, more attention is finally being turned to addressing issues with data quality. After all, no matter how much potential a particular model has, that potential cannot be realized without a good dataset to learn from. Image classification is a common task for machine learning models, and these models suffer from a particular type of data problem called co-occurrence bias. Co-occurrence bias can cause irrelevant details to get the attention of a machine learning model, leading to incorrect predictions. For example, if a dataset used to train an object recognition model only contains images of boats in the ocean, the model may start classifying anything related to the ocean, such as beaches or waves, as boats.

You’ve just spent the last month pouring over numbers, sweating the details, and making sure that everything is in order. The client is happy, the boss is happy, and life is good. But then you find…

What you're seeing in this video from graphics processing firm Nvidia is the result of two algorithmic adversaries trying to one-up each other. Working from a photo database of 30,000 celebrity faces, the two algorithms learned about different details, like beards and jewelry, that make a face look real to the human eye, and then engaged in a rapid-fire back-and-forth process that produced amazingly realistic results. None of the good-looking folks you see are real, but you'd never know it.