SAMformer: Unlocking the Potential of Transformers in Time Series Forecasting with Sharpness-Aware Minimization and Channel-Wise Attention
This work addresses a critical bottleneck in time series forecasting for researchers and practitioners by improving transformer efficiency and performance, though it is incremental as it builds on existing transformer and optimization methods.
The paper tackled the underperformance of transformers in multivariate long-term time series forecasting compared to linear baselines, identifying attention mechanisms as a cause of low generalization and proposing SAMformer, which uses sharpness-aware minimization to achieve state-of-the-art results on real-world datasets, matching the performance of the large foundation model MOIRAI with far fewer parameters.
Transformer-based architectures achieved breakthrough performance in natural language processing and computer vision, yet they remain inferior to simpler linear baselines in multivariate long-term forecasting. To better understand this phenomenon, we start by studying a toy linear forecasting problem for which we show that transformers are incapable of converging to their true solution despite their high expressive power. We further identify the attention of transformers as being responsible for this low generalization capacity. Building upon this insight, we propose a shallow lightweight transformer model that successfully escapes bad local minima when optimized with sharpness-aware optimization. We empirically demonstrate that this result extends to all commonly used real-world multivariate time series datasets. In particular, SAMformer surpasses current state-of-the-art methods and is on par with the biggest foundation model MOIRAI while having significantly fewer parameters. The code is available at https://github.com/romilbert/samformer.