TFGformer: Multivariate Time Series Forecasting via Time-Frequency Graph Learning and Covariate Fusion

The supplied arXiv metadata names the paper TFGformer, describing time-frequency graph learning and covariate fusion for long-horizon multivariate forecasting. However, the supplied abstract describes a different framework, CrossRAG: Shape-Aware Memory (SAM) with RevIN for magnitude-robust retrieval, Future-Consistent Contrastive (FCC) learning, and Cross-Attention Temporal Fusion (CATF). It claims consistent gains over parametric baselines and prior retrieval-augmented forecasting methods on seven public benchmarks. Because the title and abstract are inconsistent, the technical contribution and evaluation should be verified against the full paper.