Shijian Deng Kai Wang Tianyu Yang Harsh Singh Yapeng Tian

Abstract

Recent advancements in self-improvement for Large Language Models (LLMs) haveefficiently enhanced model capabilities without significantly increasing costs,particularly in terms of human effort. While this area is still relativelyyoung, its extension to the multimodal domain holds immense potential forleveraging diverse data sources and developing more general self-improvingmodels. This survey is the first to provide a comprehensive overview ofself-improvement in Multimodal LLMs (MLLMs). We provide a structured overviewof the current literature and discuss methods from three perspectives: 1) datacollection, 2) data organization, and 3) model optimization, to facilitate thefurther development of self-improvement in MLLMs. We also include commonly usedevaluations and downstream applications. Finally, we conclude by outlining openchallenges and future research directions.