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

2025-07-28

AI Summary

To address performance degradation in time-series classification (TSC) caused by inter-class similarity, multimodal class distributions, and noise, this paper proposes CDNet—a novel framework integrating contrastive learning with diffusion models. Its core innovation is a cross-sample, cross-class contrastive diffusion mechanism: leveraging reverse diffusion to implicitly generate discriminative positive and negative samples, thereby achieving implicit denoising and comprehensive mode coverage. Additionally, an uncertainty-weighted composite loss function is introduced to enhance training robustness. Built upon a lightweight CNN backbone, CDNet achieves significant improvements over state-of-the-art methods on the UCR Archive and synthetic multimodal/noisy datasets. It demonstrates exceptional performance under high inter-class similarity, strong noise contamination, and multimodal settings, validating its effectiveness and generalizability in modeling complex time-series distributions.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Diffusion Models for Vision
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Web Mining and Content Analysis: Content-based information diffusion
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

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.

Problem

• Enhances time series classification under class similarity
• Improves performance with multimodal and noisy data
• Generates informative samples via contrastive diffusion process

Innovation

• Generates contrastive samples via diffusion process
• Uses CNN-based reverse diffusion steps
• Employs uncertainty-weighted composite loss