Model-agnostic Retrieval-Augmented Extended Forecasting for time series

📅 2026-08-14
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
This study addresses the high fine-tuning costs associated with adapting pretrained models for short-horizon time series forecasting by proposing RAEF, a model-agnostic framework. RAEF innovatively employs input-space retrieval coupled with a structure-preserving concatenation aggregation mechanism to enable efficient adaptation without parameter updates. Establishing a new paradigm for retrieval-augmented generation in temporal domains, this approach supports zero-shot forecasting while effectively mitigating computational overhead. Experimental results demonstrate that RAEF surpasses RAF in both accuracy and inference efficiency, achieving performance comparable to full fine-tuning. Consequently, this work provides an effective and lightweight solution for time series forecasting, significantly reducing the resource demands typically associated with adapting large pretrained models to short-history scenarios.
📝 Abstract
Time series forecasting with pretrained foundation models has demonstrated strong zero-shot capabilities. However, achieving optimal performance on time series with short or negligible historical data in domain-specific applications typically requires adaptation via either fine-tuning or RAG. While fine-tuning is effective, it incurs substantial computational costs. This work explores RAG within univariate time series (Retrieval Augmented Generation) as a more efficient alternative, in particular RAF (Retrieval Augmented Forecasting), and introduces RAEF (Retrieval-Augmented Extended Forecasting), a model-agnostic method built upon RAF. RAEF incorporates key refinements to the retrieval and aggregation mechanisms: (1) direct retrieval in input-space rather than embedding-space, reducing inference overhead, and (2) concatenation-based aggregation that preserves temporal structure instead of averaging. Empirical evaluation across multiple benchmark datasets demonstrates that RAEF outperforms RAF in both accuracy and inference overhead. Furthermore, comprehensive comparisons with zero-shot and fine-tuned foundation models show that RAEF achieves competitive or superior performance to fine-tuning while avoiding its computational burden, establishing it as a practical and scalable approach for domain adaptation in time series forecasting.
Problem

Research questions and friction points this paper is trying to address.

Time Series Forecasting
Retrieval-Augmented Generation
Domain Adaptation
Pretrained Foundation Models
Computational Efficiency
Innovation

Methods, ideas, or system contributions that make the work stand out.

Retrieval-Augmented Extended Forecasting
Model-agnostic
Input-space retrieval
Concatenation-based aggregation
Time series forecasting