CompoDistill: Attention Distillation for Compositional Reasoning in Multimodal LLMs

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Kim, Jiwan, Kim, Kibum, Seo, Sangwoo, Park, Chanyoung
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911577776586752
author Kim, Jiwan
Kim, Kibum
Seo, Sangwoo
Park, Chanyoung
author_facet Kim, Jiwan
Kim, Kibum
Seo, Sangwoo
Park, Chanyoung
contents Recently, efficient Multimodal Large Language Models (MLLMs) have gained significant attention as a solution to their high computational complexity, making them more practical for real-world applications. In this regard, the knowledge distillation (KD) approach has emerged as a promising alternative, which transfers the rich visual and linguistic knowledge from a larger model (teacher) to a smaller model (student). However, we observe that existing KD methods struggle to effectively distill the teacher MLLM's rich visual perception abilities to the student, a challenge that has been largely overlooked in previous studies. Through a systematic analysis, we identify visual attention misalignment between student and teacher as the main cause of this issue. Based on this insight, we propose CompoDistill, a novel KD framework that explicitly aligns the student's visual attention with that of the teacher to enhance the student's visual perception abilities. Our extensive experiments show that CompoDistill significantly improves performance on compositional reasoning tasks that require visual perception abilities while maintaining strong performance on visual question answering tasks, as done in existing studies. Furthermore, CompoDistill demonstrates effectiveness with a more advanced backbone, highlighting its generalizability.
format Preprint
id arxiv_https___arxiv_org_abs_2510_12184
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CompoDistill: Attention Distillation for Compositional Reasoning in Multimodal LLMs
Kim, Jiwan
Kim, Kibum
Seo, Sangwoo
Park, Chanyoung
Computer Vision and Pattern Recognition
Artificial Intelligence
Recently, efficient Multimodal Large Language Models (MLLMs) have gained significant attention as a solution to their high computational complexity, making them more practical for real-world applications. In this regard, the knowledge distillation (KD) approach has emerged as a promising alternative, which transfers the rich visual and linguistic knowledge from a larger model (teacher) to a smaller model (student). However, we observe that existing KD methods struggle to effectively distill the teacher MLLM's rich visual perception abilities to the student, a challenge that has been largely overlooked in previous studies. Through a systematic analysis, we identify visual attention misalignment between student and teacher as the main cause of this issue. Based on this insight, we propose CompoDistill, a novel KD framework that explicitly aligns the student's visual attention with that of the teacher to enhance the student's visual perception abilities. Our extensive experiments show that CompoDistill significantly improves performance on compositional reasoning tasks that require visual perception abilities while maintaining strong performance on visual question answering tasks, as done in existing studies. Furthermore, CompoDistill demonstrates effectiveness with a more advanced backbone, highlighting its generalizability.
title CompoDistill: Attention Distillation for Compositional Reasoning in Multimodal LLMs
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2510.12184