A Contrastive Diffusion-based Network (CDNet) for Time Series Classification
This work addresses robustness issues in time series classification for applications dealing with noisy or complex data, representing an incremental advance by building on existing diffusion and contrastive learning methods.
The paper tackles performance degradation in deep learning models for time series classification under challenging conditions like class similarity, multimodal distributions, and noise by proposing CDNet, a Contrastive Diffusion-based Network that enhances classifiers through sample generation, resulting in significant improvements on the UCR Archive and simulated datasets.
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.