aiqbench analysis
This research identifies "multimodal saturation," where adding a third data modality degrades performance in standard self-supervised learning frameworks like VICReg due to geometric alignment issues. The authors propose Riemannian VICReg, which utilizes negative-curvature product factors to correct this geometry and successfully recovers the benefits of trimodal training in previously failing scenarios.
On the Geometry of Multimodal Saturation: Riemannian VICReg
arXiv:2610.06096v1 Announce Type: cross Abstract: In self-supervised learning, a third modality should improve, or at least preserve, performance. Across nine image-text-tabular datasets, we show that it instead harms performance: the trimodal model underperforms its own best bimodal subset in 55.6% of paired runs under VICReg. The same failure occurs in 51.1% of paired runs under SimSiam. We call this failure multimodal saturation. We propose that the failure lies in the alignment geometry. Riemannian VICReg (R-VICReg) generalizes classical VICReg: it aligns views by squared geodesic distance on learnable negative-curvature product factors and recovers VICReg exactly as curvature vanishes. Over the same 45 paired runs, R-VICReg raises the probability that the third modality helps from 44.4% to 64.4%, with gains concentrated where VICReg saturates.
Abstract:In self-supervised learning, a third modality should improve, or at least preserve, performance. Across nine image-text-tabular datasets, we show that it instead harms performance: the trimodal model underperforms its own best bimodal subset in 55.6% of paired runs under VICReg. The same failure occurs in 51.1% of paired runs under SimSiam. We call this failure multimodal saturation. We propose that the failure lies in the alignment geometry. Riemannian VICReg (R-VICReg) generalizes classical VICReg: it aligns views by squared geodesic distance on learnable negative-curvature product factors and recovers VICReg exactly as curvature vanishes. Over the same 45 paired runs, R-VICReg raises the probability that the third modality helps from 44.4% to 64.4%, with gains concentrated where VICReg saturates.
17 pages, 6 figures. Accepted to the Proceedings Track of the NeurIPS 2026 Workshop on Symmetry and Geometry in Neural Representations (NeurReps)
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2610.06096 [cs.LG]
(or arXiv:2610.06096v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.06096 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 · 236 words
Related coverage
Beyond Unimodal Bases: Pullback Geometry for Multimodal Data
arXiv · Oct 2, 2026
RGSQ: Riemannian Geometry-Sensitive Quantization for Large Vision-Language Models
arXiv · Sep 24, 2026
Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery
arXiv · Jul 23, 2026
Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery
arXiv · Jun 9, 2026