




















Abstract:Context-aided forecasting (CAF) holds promise for integrating domain knowledge and forward-looking information, enabling AI systems to surpass traditional statistical methods. However, recent empirical studies reveal a puzzling gap: multimodal models often fail to outperform their unimodal counterparts. We hypothesize that this underperformance stems from poor context quality in existing datasets, as verification is challenging. To address these limitations, we introduce a semi-synthetic data augmentation method that generates contexts both descriptive of temporal dynamics and verifiably complementary to numerical histories. This approach enables massive-scale dataset creation, resulting in CAF-7M, a corpus of 7 million context-augmented time series windows, including a rigorously verified test set. We demonstrate that semi-synthetic pre-training transfers effectively to real-world evaluation, and show clear evidence of context utilization. Our results suggest that dataset quality, rather than architectural limitations, has been the primary bottleneck in context-aided forecasting.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2603.12451 [cs.LG] |
| (or arXiv:2603.12451v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2603.12451 arXiv-issued DOI via DataCite |
From: Étienne Marcotte [view email]
[v1]
Thu, 12 Mar 2026 21:05:33 UTC (903 KB)
[v2]
Mon, 16 Mar 2026 19:46:16 UTC (903 KB)
[v3]
Wed, 22 Apr 2026 15:43:58 UTC (908 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理,仅供阅读参考。 原文来自 — 版权归原作者所有。