TS-RAG: Retrieval Augmented Generation for Time Series Forecasting
The paper adapts RAG to forecasting by retrieving similar time-series sequences and fusing them through dedicated reference tokens.
The authors argue that simply appending retrieved sequences, as LLM prompts often do, is a poor fit for smaller time-series models. TS-RAG is designed to combine the input series with retrieved references so the model can capture more complex temporal dynamics. They report consistent state-of-the-art results across several real-world forecasting benchmarks. ArXiv · AI/CL/LG's note
The authors argue that simply appending retrieved sequences, as LLM prompts often do, is a poor fit for smaller time-series models. TS-RAG is designed to combine the input series with retrieved references so the model can capture more complex temporal dynamics. They report consistent state-of-the-art results across several real-world forecasting benchmarks. ArXiv · AI/CL/LG's note
score 4