COSCO: A Sharpness-Aware Training Framework for Few-shot Multivariate Time Series Classification
Jesus Barreda, Ashley Gomez, Ruben Puga, Kaixiong Zhou, Li Zhang
TL;DR
The paper tackles few-shot multivariate time series classification by addressing generalization gaps caused by sharp loss landscapes in data-scarce regimes. It introduces COSCO, a framework that combines Sharpness-Aware Minimization (SAM) with a Prototypical loss to produce robust embeddings and flatter minima without requiring test data during training. Empirical results on 21 UEA datasets across 1- and 10-shot settings show COSCO achieving top or near-top accuracy and the best average ranking, with ablation studies confirming the contribution of both SAM and the prototypical loss. The approach offers a practical pathway to reliable MTSC under limited labeled data, with code available for reproducibility and extension.
Abstract
Multivariate time series classification is an important task with widespread domains of applications. Recently, deep neural networks (DNN) have achieved state-of-the-art performance in time series classification. However, they often require large expert-labeled training datasets which can be infeasible in practice. In few-shot settings, i.e. only a limited number of samples per class are available in training data, DNNs show a significant drop in testing accuracy and poor generalization ability. In this paper, we propose to address these problems from an optimization and a loss function perspective. Specifically, we propose a new learning framework named COSCO consisting of a sharpness-aware minimization (SAM) optimization and a Prototypical loss function to improve the generalization ability of DNN for multivariate time series classification problems under few-shot setting. Our experiments demonstrate our proposed method outperforms the existing baseline methods. Our source code is available at: https://github.com/JRB9/COSCO.