Meta makes AI video computing costs plummet by 95%, and AI can guess the original picture even if the picture is half covered

Meta researchers have published a series of new papers on MAE (masked auto-encoder), a system that uses SSL technology (self-supervised learning) to predict missing parts of data and restore missing text, images, video and audio.

The general principle of the MAE system for restoring different types of files is to predict what is missing based on existing information, and then compensate for it with other data.

With this technology, AI may be able to automatically annotate data (ground truth) instead of having to annotate it manually. This means that the learning efficiency of AI models has been greatly improved, which may bring new ideas for the future development of AI models.

I. The essence of intelligence is predictive ability, and SSL technology can improve AI intelligence

The MAE system uses SSL (Self-supervised Learning), a technique in which the annotations used for machine learning originate from the data itself, rather than from manual annotations.