ReCast: Reliability-aware Codebook Assisted Lightweight Time Series Forecasting

TL;DR

ReCast addresses non-stationary time series forecasting by leveraging local recurring shapes through patch-wise quantization into a learnable codebook. It employs a dual-path forecasting scheme—one path models regular local patterns via discrete embeddings, while a residual path recovers irregular fluctuations—yielding a lightweight yet robust predictor with predictions formed as $\\hat{\\mathbf{Y}} = \\\mathbf{Y}_q + \\\mathbf{Y}_r$. A key novelty is the reliability-aware, incremental codebook updating mechanism, which fuses three reliability signals through a distributionally robust optimization (DRO) formulation to adapt to distribution shifts: $\\hat{w}_k^t = -\\gamma \\log \\sum_i \\exp(-z_{k,i}^t/\\gamma)$. Extensive experiments on eight real-world datasets demonstrate state-of-the-art accuracy with reduced computational cost, confirming ReCast’s effectiveness and practicality for resource-constrained forecasting tasks.

Abstract

Time series forecasting is crucial for applications in various domains. Conventional methods often rely on global decomposition into trend, seasonal, and residual components, which become ineffective for real-world series dominated by local, complex, and highly dynamic patterns. Moreover, the high model complexity of such approaches limits their applicability in real-time or resource-constrained environments. In this work, we propose a novel \textbf{RE}liability-aware \textbf{C}odebook-\textbf{AS}sisted \textbf{T}ime series forecasting framework (\textbf{ReCast}) that enables lightweight and robust prediction by exploiting recurring local shapes. ReCast encodes local patterns into discrete embeddings through patch-wise quantization using a learnable codebook, thereby compactly capturing stable regular structures. To compensate for residual variations not preserved by quantization, ReCast employs a dual-path architecture comprising a quantization path for efficient modeling of regular structures and a residual path for reconstructing irregular fluctuations. A central contribution of ReCast is a reliability-aware codebook update strategy, which incrementally refines the codebook via weighted corrections. These correction weights are derived by fusing multiple reliability factors from complementary perspectives by a distributionally robust optimization (DRO) scheme, ensuring adaptability to non-stationarity and robustness to distribution shifts. Extensive experiments demonstrate that ReCast outperforms state-of-the-art (SOTA) models in accuracy, efficiency, and adaptability to distribution shifts.

Figures (5)

• Figure 1: ReCast overview. It comprises patch-wise quantization, dual-path forecasting, and codebook construction and updating.
• Figure 2: Illustration of reliability-aware scoring, showing three scoring factors and their fusion via distributionally robust optimization (DRO).
• Figure 3: Computational efficiency of ReCast on the ECL dataset (horizon = 720).
• Figure 4: Visualization of codebook evolution and cluster assignments across epochs.
• Figure 5: Performance comparison under varying hyper-parameters.