Facebook Develops SEER, a Self-Supervised AI Model for Computer Vision
Facebook introduces SEER, a billion-parameter computer vision model that learns from random, unlabeled images to achieve state-of-the-art performance. This self-supervised approach brings AI closer to human-like learning without relying on curated datasets.
Facebook announces a new artificial intelligence model called SEER that achieves state-of-the-art results on computer vision benchmarks. Unlike traditional models that require carefully labeled datasets, this system trains on a billion random, unlabeled images from the internet by exposing the relationships between different parts of the data.
This self-supervised approach represents a major step toward human-level intelligence, as the AI learns to recognize visual concepts without human annotation. To accomplish this complex task, Facebook combines a large convolutional network inspired by the human visual cortex with an algorithm called SwAV that groups similar images together through clustering techniques.
By leveraging these new techniques, SEER overcomes the challenge of understanding how pixels relate to concepts across varying images. The model requires significantly less training time than previous state-of-the-art systems, proving that AI can effectively make inferences from unorganized information rather than relying on meticulously curated data.