This research provides a systematic taxonomy of connector components that enable large language models to process multiple data types (text, images, audio, etc.) simultaneously.

• Identifies the critical role of connectors in bridging diverse modalities and enhancing MLLM performance
• Presents a structured taxonomy of connector designs and architectures
• Analyzes the evolution and current state of connector technologies
• Highlights engineering implications for developing more powerful multi-modal AI systems

For engineers and AI developers, this research offers valuable insights into designing more effective cross-modal integration components, a critical factor in building next-generation AI systems that can seamlessly understand and process multiple data formats.

Connector-S: A Survey of Connectors in Multi-modal Large Language Models