CONTINUAL LEARNING WITH MOE ADAPTERS

Summary from the TLDR AI newsletter

Researchers have developed a new framework to help vision-language models learn continuously without forgetting previous knowledge using a system that expands the model with special adapters for new tasks.