Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.05184 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909828575657984 |
|---|---|
| author | Lu, Jianglin Wang, Hailing Xu, Yi Wang, Yizhou Yang, Kuo Fu, Yun |
| author_facet | Lu, Jianglin Wang, Hailing Xu, Yi Wang, Yizhou Yang, Kuo Fu, Yun |
| contents | Foundation models learn highly transferable representations through large-scale pretraining on diverse data. An increasing body of research indicates that these representations exhibit a remarkable degree of similarity across architectures and modalities. In this survey, we investigate the representation potentials of foundation models, defined as the latent capacity of their learned representations to capture task-specific information within a single modality while also providing a transferable basis for alignment and unification across modalities. We begin by reviewing representative foundation models and the key metrics that make alignment measurable. We then synthesize empirical evidence of representation potentials from studies in vision, language, speech, multimodality, and neuroscience. The evidence suggests that foundation models often exhibit structural regularities and semantic consistencies in their representation spaces, positioning them as strong candidates for cross-modal transfer and alignment. We further analyze the key factors that foster representation potentials, discuss open questions, and highlight potential challenges. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_05184 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Representation Potentials of Foundation Models for Multimodal Alignment: A Survey Lu, Jianglin Wang, Hailing Xu, Yi Wang, Yizhou Yang, Kuo Fu, Yun Artificial Intelligence Foundation models learn highly transferable representations through large-scale pretraining on diverse data. An increasing body of research indicates that these representations exhibit a remarkable degree of similarity across architectures and modalities. In this survey, we investigate the representation potentials of foundation models, defined as the latent capacity of their learned representations to capture task-specific information within a single modality while also providing a transferable basis for alignment and unification across modalities. We begin by reviewing representative foundation models and the key metrics that make alignment measurable. We then synthesize empirical evidence of representation potentials from studies in vision, language, speech, multimodality, and neuroscience. The evidence suggests that foundation models often exhibit structural regularities and semantic consistencies in their representation spaces, positioning them as strong candidates for cross-modal transfer and alignment. We further analyze the key factors that foster representation potentials, discuss open questions, and highlight potential challenges. |
| title | Representation Potentials of Foundation Models for Multimodal Alignment: A Survey |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2510.05184 |