Divide and Contrast: Learning Robust Temporal Features without Augmentation

2026-05-20

AI Summary

Existing self-supervised representation learning methods for time series are often limited by high computational costs, reliance on task-specific data augmentations, or strong assumptions about temporal dynamics. This work proposes Di-COT, a novel framework that randomly partitions each time window into overlapping sub-blocks and leverages augmentation-free contrastive learning to capture local semantic structures while effectively mitigating false-positive interference. Di-COT introduces a scalable contrastive loss independent of sequence length, significantly enhancing both training efficiency and generalization capability. Evaluated across six real-world datasets and the UCR/UEA benchmarks, Di-COT achieves state-of-the-art performance in classification, clustering, kNN retrieval, and cross-dataset transfer tasks, while substantially reducing training time.

Abstract

Self-supervised learning for time-series representation aims to reduce reliance on labeled data while maintaining strong downstream performance, yet many existing approaches incur high computational costs or rely on assumptions that do not hold across diverse temporal dynamics. In this work, we introduce Divide and Contrast (Di-COT), an unsupervised framework that avoids data augmentation and multiple encoder passes by contrasting informative substructures within a window rather than individual timesteps. Di-COT stochastically partitions each window into a small number of overlapping sub-blocks per iteration, enabling efficient and meaningful contrast while mitigating false positives during temporal transitions. To further improve scalability, we adopt a contrastive objective whose computation depends on the batch size and the number of sub-blocks, making loss computation independent of sequence length. Extensive experiments on six large-scale real-world datasets, as well as the UCR and UEA benchmarks, demonstrate that Di-COT learns semantically structured and transferable representations, achieving state-of-the-art performance on classification, clustering, $k$NN, and cross-dataset transfer, while substantially reducing training time. The source code is publicly available at https://github.com/sfi-norwai/Di-COT.

Problem

• self-supervised learning
• time-series representation
• data augmentation
• temporal dynamics
• computational cost

Innovation

• self-supervised learning
• time-series representation
• data augmentation-free
• contrastive learning
• temporal substructures