AI's multi-view wave is coming, and it will be powerful

Multi-view, a approach that links different data types as views of the same object, is expected to become central as neural networks take on more modalities, according to the article. Current systems such as Meta's SeamlessM4T process text and speech together but do not explicitly link them as facets of one object. In a paper posted on arXiv in April, Ravid Shwartz-Ziv and Yann LeCun argue that the challenge is separating relevant from irrelevant information in an information bottleneck as views multiply, and that the traditional two-view assumption may be too restrictive.