aiqbench analysis
This research introduces CrossRAG, a novel framework leveraging retrieval-augmented generation (RAG) to enhance multivariate time series forecasting by dynamically accessing external historical patterns from heterogeneous IoT sensors. By integrating shape-aware memory and future-consistent contrastive learning, the method effectively addresses magnitude variations and temporal mismatches that typically hinder standard foundation models in predictive maintenance scenarios. The approach demonstrates superior performance over static parametric baselines across seven public benchmarks, offering a significant step forward for dynamic resource scheduling applications.
TFGformer: Multivariate Time Series Forecasting via Time-Frequency Graph Learning and Covariate Fusion
arXiv:2607.29459v1 Announce Type: new Abstract: Large-scale multivariate time series from heterogeneous IoT sensors demand accurate long-term forecasting for resource scheduling and predictive maintenance. While recent time series foundation models exhibit strong generalization, they rely on static parametric knowledge and lack dynamic access to external historical patterns during inference. Retrieval-Augmented Generation (RAG) offers a potential remedy, yet its application to time series forecasting is challenged by magnitude variations across heterogeneous sources and the mismatch between historical similarity and future consistency. We propose CrossRAG, a retrieval-augmented forecasting framework that integrates Shape-Aware Memory (SAM) with RevIN normalization for magnitude-robust shape-level retrieval, Future-Consistent Contrastive (FCC) learning to distinguish informative references from hard negatives with similar history but divergent futures, and Cross-Attention Temporal Fusion (CATF) to fuse retrieved historical--future reference pairs into the backbone's representations at the representation level. Experiments on seven public benchmarks show that CrossRAG consistently outperforms both parametric-only baselines and existing retrieval-augmented forecasting methods.
Abstract:Large-scale multivariate time series from heterogeneous IoT sensors demand accurate long-term forecasting for resource scheduling and predictive maintenance. While recent time series foundation models exhibit strong generalization, they rely on static parametric knowledge and lack dynamic access to external historical patterns during inference. Retrieval-Augmented Generation (RAG) offers a potential remedy, yet its application to time series forecasting is challenged by magnitude variations across heterogeneous sources and the mismatch between historical similarity and future consistency. We propose CrossRAG, a retrieval-augmented forecasting framework that integrates Shape-Aware Memory (SAM) with RevIN normalization for magnitude-robust shape-level retrieval, Future-Consistent Contrastive (FCC) learning to distinguish informative references from hard negatives with similar history but divergent futures, and Cross-Attention Temporal Fusion (CATF) to fuse retrieved historical--future reference pairs into the backbone's representations at the representation level. Experiments on seven public benchmarks show that CrossRAG consistently outperforms both parametric-only baselines and existing retrieval-augmented forecasting methods.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2607.29459 [cs.LG]
(or arXiv:2607.29459v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2607.29459 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 · 244 words
Related coverage
SpikF-GO: Spiking Fourier Graph Operators for Multivariate Time Series Forecasting
arXiv · Jun 15, 2026
TiWeaver: Unified Temporal Dynamics Modeling via Contextual Patching
arXiv · Jun 3, 2026
CoRe: Coherence and Relational Alignment for Multivariate Time Series Forecasting
arXiv · Sep 18, 2026
[AINews] Fal’s H3 Max Live breaks the infinite videogen barrier
Latent Space · Sep 1, 2026