aiqbench analysis
This research addresses a critical inefficiency in current time series forecasting by demonstrating that naive multimodal fusion often degrades performance rather than improving it. The authors introduce the Controlled Fusion Adapter (CFA), which utilizes low-rank adapters to filter out irrelevant textual noise before integrating it into temporal models, offering a generalizable solution across thousands of experiments. This work provides essential architectural guidance for developers seeking robust AI systems that effectively leverage auxiliary data without compromising core forecasting accuracy.
Rethinking Multimodal Fusion for Time Series: Text Modalities Need Constrained Fusion
arXiv:2603.22372v2 Announce Type: replace Abstract: Recent advances in multimodal learning have motivated the integration of auxiliary modalities such as text or vision into time series (TS) forecasting. However, most existing methods provide limited gains, often improving performance only in specific datasets or relying on architecture-specific designs that limit generalization. In this paper, we show that multimodal models with naive fusion strategies (e.g., simple addition or concatenation) often underperform unimodal TS models, which we attribute to the uncontrolled integration of auxiliary modalities which may introduce irrelevant information. Motivated by this observation, we explore various constrained fusion methods designed to control such integration and find that they consistently outperform naive fusion methods. Furthermore, we propose Controlled Fusion Adapter (CFA), a simple plug-in method that enables controlled cross-modal interactions without modifying the TS backbone, integrating only relevant textual information aligned with TS dynamics. CFA employs low rank adapters to filter irrelevant textual information before fusing it into temporal representations. We conduct over 20K experiments across various datasets and TS/text models, demonstrating the effectiveness of the constrained fusion methods. Code is available at: https://github.com/seunghan96/cfa.
Abstract:Recent advances in multimodal learning have motivated the integration of auxiliary modalities such as text or vision into time series (TS) forecasting. However, most existing methods provide limited gains, often improving performance only in specific datasets or relying on architecture-specific designs that limit generalization. In this paper, we show that multimodal models with naive fusion strategies (e.g., simple addition or concatenation) often underperform unimodal TS models, which we attribute to the uncontrolled integration of auxiliary modalities which may introduce irrelevant information. Motivated by this observation, we explore various constrained fusion methods designed to control such integration and find that they consistently outperform naive fusion methods. Furthermore, we propose Controlled Fusion Adapter (CFA), a simple plug-in method that enables controlled cross-modal interactions without modifying the TS backbone, integrating only relevant textual information aligned with TS dynamics. CFA employs low rank adapters to filter irrelevant textual information before fusing it into temporal representations. We conduct over 20K experiments across various datasets and TS/text models, demonstrating the effectiveness of the constrained fusion methods. Code is available at: this https URL.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2603.22372 [cs.LG]
(or arXiv:2603.22372v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2603.22372 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.
Full text under the publisher's open licence · 272 words
Related coverage
Rethinking Self-Evolution: A Constrained Exploration-Exploitation Process for Mitigating Skill Overfitting
arXiv · Jul 31, 2026
Power up your AI infrastructure! A first look at the Smart Systems Stage agenda at TechCrunch Disrupt 2026
TechCrunch · Jul 27, 2026
Rethinking the Information Bottleneck: Structured Decomposition under Label-Induced Partitions
arXiv · Oct 2, 2026
HazardArena: Evaluating Semantic Safety in Vision-Language-Action Models
arXiv · Sep 23, 2026