作者位次 #3/5ACL 2026Poster· Natural Language Processing
From Experts to Bases: Orthogonal Subspace Mixture for Continual Multimodal Instruction Tuning
Pei Chen · Xilai Wang · Shiqixu · Zejian Li · Lingyun Sun
摘要
Multimodal Continual Instruction Tuning (MCIT) is essential for adapting Multimodal Large Language Models (MLLMs) to dynamic data streams, yet preventing catastrophic forgetting remains a major challenge. Existing parameter-efficient approaches often face a dilemma: fixed architectures suffer fro...