Scaleformer: Iterative Multi-scale Refining Transformers for Time Series Forecasting

TL;DR

Scaleformer introduces an iterative multi-scale refining framework for transformer-based time series forecasting, significantly enhancing accuracy with minimal extra computation by refining predictions at multiple scales.

Contribution

We propose a novel multi-scale framework with shared weights and normalization for transformer models, improving forecasting accuracy across datasets.

Findings

Performance improvements of 5.5% to 38.5% across datasets.

Effective architecture adaptations demonstrated via ablation studies.

Outperforms baseline transformer models on public datasets.

Abstract

The performance of time series forecasting has recently been greatly improved by the introduction of transformers. In this paper, we propose a general multi-scale framework that can be applied to the state-of-the-art transformer-based time series forecasting models (FEDformer, Autoformer, etc.). By iteratively refining a forecasted time series at multiple scales with shared weights, introducing architecture adaptations, and a specially-designed normalization scheme, we are able to achieve significant performance improvements, from 5.5% to 38.5% across datasets and transformer architectures, with minimal additional computational overhead. Via detailed ablation studies, we demonstrate the effectiveness of each of our contributions across the architecture and methodology. Furthermore, our experiments on various public datasets demonstrate that the proposed improvements outperform their corresponding baseline counterparts. Our code is publicly available in https://github.com/BorealisAI/scaleformer.