analiza aiqbench
This research introduces SP-CRL, a method designed to mitigate semantic leakage in vision-language models by constructing purer conditional subspaces through curvature-based truncation and noise subspace projection. By explicitly disentangling coupled semantics, the approach significantly improves performance on customized clustering, few-shot classification, and retrieval tasks without requiring explicit retraining of the base VLMs. The work advances the theoretical understanding of representation learning within large multimodal models, offering a robust solution for extracting criterion-specific features in complex embedding spaces.
Semantic Purification for Conditional Representation Learning
arXiv:2602.05464v2 Announce Type: replace-cross Abstract: Conditional representation learning aims to extract criterion-specific features for customized tasks. Recent methods construct conditional subspaces spanned by criterion-specific text bases in the embedding space of vision-language models (VLMs). Image embeddings are then projected onto these subspaces to obtain conditional representations. However, since VLMs are not explicitly trained to disentangle semantics associated with different criteria, the corresponding conditional subspaces remain coupled. This coupling induces semantic leakage during projection, thereby degrading the semantic purity of conditional representations. To suppress semantic leakage, we propose Semantic Purification for Conditional Representation Learning (SP-CRL). Specifically, SP-CRL first decomposes the original text basis and performs curvature-based adaptive truncation on the resulting basis vectors to construct a purer conditional subspace. It then identifies an appropriate noise subspace and projects image embeddings onto its null space to remove irrelevant semantic components. Extensive experiments across customized clustering, customized few-shot classification, and customized retrieval tasks demonstrate that SP-CRL achieves state-of-the-art performance with superior generalization.
Abstract:Conditional representation learning aims to extract criterion-specific features for customized tasks. Recent methods construct conditional subspaces spanned by criterion-specific text bases in the embedding space of vision-language models (VLMs). Image embeddings are then projected onto these subspaces to obtain conditional representations. However, since VLMs are not explicitly trained to disentangle semantics associated with different criteria, the corresponding conditional subspaces remain coupled. This coupling induces semantic leakage during projection, thereby degrading the semantic purity of conditional representations. To suppress semantic leakage, we propose Semantic Purification for Conditional Representation Learning (SP-CRL). Specifically, SP-CRL first decomposes the original text basis and performs curvature-based adaptive truncation on the resulting basis vectors to construct a purer conditional subspace. It then identifies an appropriate noise subspace and projects image embeddings onto its null space to remove irrelevant semantic components. Extensive experiments across customized clustering, customized few-shot classification, and customized retrieval tasks demonstrate that SP-CRL achieves state-of-the-art performance with superior generalization.
Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2602.05464 [cs.AI]
(or arXiv:2602.05464v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2602.05464 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 255 cuvinte
Subiecte conexe
DVLA-RL++: Dual-Level Vision-Language Alignment with Reinforcement Learning Gating for Few-Shot Learning
arXiv · 9 oct. 2026
AlignAudio: A Unified Dual-Alignment Framework for Noise-Robust Speech-Conditioned Generation
arXiv · 15 oct. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
Multimodal Representation Learning Conditioned on Semantic Relations
arXiv · 6 oct. 2026