DeepMind’s GATS: A Novel Module for Seamless Integration of Multimodal Foundation Models

In a new paper GATS: Gather-Attend-Scatter, a Google DeepMind research team introduces Gather-Attend-Scatter (GATS), a pioneering module designed to seamlessly combine pretrained foundation models—whether trainable or frozen—into larger multimodal networks.

ai artificial intelligence deepmind deepmind research deep-neural-networks foundation foundation model gather google google deepmind integration machine learning machine learning & data science ml multimodal multimodal model networks novel paper research research team team technology