As per a blog post published by Facebook AI, one billion Instagram photos were used to train its AI-based SEER algorithm (an abbreviation of self-supervised) which recognizes images on its own. Algorithms are usually trained as per datasets that are already categorized by humans and are labeled. However, Instagram pictures were fed into this algorithm without any such labeling.
The SEER algorithm identified the images with high accuracy, Facebook reported. “SEER outperformed the most advanced, state-of-the-art self-supervised systems, reaching 84.2 percent top-1 accuracy on ImageNet. SEER also outperformed state-of-the-art supervised models on downstream tasks, including low-shot, object detection, segmentation, and image classification.”, the post said.
Facebook also said that while a similar technique has been used for algorithms dealing with processing language, images pose a rather complex challenge. However, the company said that it is sharing details to democratize self-supervised learning.
“Making progress on a challenge this broad and deep requires the open exchange of ideas among diverse minds in the field. We remain committed to the principles of open science and hope that this brings the field significantly closer to building machines that understand the visual world, as well as people, do.”, the post added.
The post also said that training algorithms with massive datasets could help eliminate bias due to people’s physical characteristics.