A Contrastive Diffusion-based Network (CDNet) for Time Series Classification

TL;DR

CDNet introduces a contrastive diffusion approach to enhance time series classification, effectively handling noisy, similar, and multimodal data by generating informative samples and improving classifier robustness.

Contribution

The paper proposes CDNet, a novel diffusion-based network that learns sample transitions for better class separation and robustness in challenging time series data.

Findings

Significantly outperforms SOTA classifiers on UCR datasets.

Improves robustness under noisy and multimodal conditions.

Effectively generates informative positive and negative samples.

Abstract

Deep learning models are widely used for time series classification (TSC) due to their scalability and efficiency. However, their performance degrades under challenging data conditions such as class similarity, multimodal distributions, and noise. To address these limitations, we propose CDNet, a Contrastive Diffusion-based Network that enhances existing classifiers by generating informative positive and negative samples via a learned diffusion process. Unlike traditional diffusion models that denoise individual samples, CDNet learns transitions between samples--both within and across classes--through convolutional approximations of reverse diffusion steps. We introduce a theoretically grounded CNN-based mechanism to enable both denoising and mode coverage, and incorporate an uncertainty-weighted composite loss for robust training. Extensive experiments on the UCR Archive and simulated datasets demonstrate that CDNet significantly improves state-of-the-art (SOTA) deep learning classifiers, particularly under noisy, similar, and multimodal conditions.