aiqbench analysis
This research identifies a counterintuitive failure mode where multimodal models trained on complete data underperform unimodal counterparts when one input stream is missing at inference time. The authors attribute this degradation to structured rotations in parameter subspaces induced by cross-modal dependencies and propose Geodesic Unlearning, a lightweight geometric editing method that corrects these subspaces without sacrificing full-modality accuracy. By leveraging Grassmannian geometry, the approach offers a practical route to enhancing deployment-time robustness across diverse architectures like CLIP and fusion models.
Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness
arXiv:2610.04792v1 Announce Type: new Abstract: Missing modality remains a longstanding challenge in multimodal learning. Existing methods typically address this issue through modality recovery or adaptive strategies. However, they overlook models' internal cross-modal dependencies formed during multimodal training, which later impair robustness. We systematically characterize a counterintuitive deployment-time failure mode: models trained on full modalities can underperform unimodal models when one modality is missing at inference time. This pattern appears across diverse architectures, such as fusion models, CLIP-style two-tower models, and vision-language models. We show that such degradation is closely associated with learned cross-modal dependencies in the principal parameter subspaces. Multimodal training induces structured rotations of these subspaces, particularly in cross-modal interaction layers. These rotations are associated with reduced task-aligned margins and larger task-aware representation harm under missing-modality inputs. We propose Geodesic Unlearning (GU), a lightweight parameter-editing method that leverages Grassmannian subspace geometry for structured subspace correction to improve missing-modality robustness. It rotates the principal input subspace toward a unimodal reference along a geodesic path. We prove that this correction minimizes the distance to the reference within a fixed subspace-distance budget. Experiments across architectures and datasets show that GU improves performance under missing-modality inference while preserving full-modality accuracy, outperforming strong missing-modality robustness baselines. These findings support a geometric view of deployment-time missing-modality degradation and suggest localized subspace editing as a practical route for robustness correction.
Title:Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness
Abstract:Missing modality remains a longstanding challenge in multimodal learning. Existing methods typically address this issue through modality recovery or adaptive strategies. However, they overlook models' internal cross-modal dependencies formed during multimodal training, which later impair robustness. We systematically characterize a counterintuitive deployment-time failure mode: models trained on full modalities can underperform unimodal models when one modality is missing at inference time. This pattern appears across diverse architectures, such as fusion models, CLIP-style two-tower models, and vision-language models. We show that such degradation is closely associated with learned cross-modal dependencies in the principal parameter subspaces. Multimodal training induces structured rotations of these subspaces, particularly in cross-modal interaction layers. These rotations are associated with reduced task-aligned margins and larger task-aware representation harm under missing-modality inputs. We propose Geodesic Unlearning (GU), a lightweight parameter-editing method that leverages Grassmannian subspace geometry for structured subspace correction to improve missing-modality robustness. It rotates the principal input subspace toward a unimodal reference along a geodesic path. We prove that this correction minimizes the distance to the reference within a fixed subspace-distance budget. Experiments across architectures and datasets show that GU improves performance under missing-modality inference while preserving full-modality accuracy, outperforming strong missing-modality robustness baselines. These findings support a geometric view of deployment-time missing-modality degradation and suggest localized subspace editing as a practical route for robustness correction.
Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2610.04792 [cs.AI]
(or arXiv:2610.04792v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.04792 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Full text under the publisher's open licence · 330 words
Related coverage
Podcast with Venture Investor Russ Fein
Quantum Computing Report · Aug 27, 2026
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
METR · Aug 26, 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · Oct 13, 2026
Presentation: From AI Agent Demo to Production: Automated Testing and Evaluation
InfoQ · Sep 7, 2026