Multi-Task Diffusion Learning for Time Series Classification

• DOI: 10.3390/electronics13204015
• Publication Date: 2024-10-14T11:47:05Z
• Authors: Shaoqiu Zheng; Zhen Liu; Long Tian; Ling Ye; Shixin Zheng; Peng Peng; Wei Chu

Current deep learning models for time series often face challenges with generalizability in scenarios characterized by limited samples or inadequately labeled data. By tapping into the robust generative capabilities of diffusion models, which have shown success computer vision and natural language processing, we see potential improving adaptability models. However, specific application generating classification tasks remains underexplored. To bridge this gap, introduce MDGPS model, incorporates multi-task gradient-free patch search (MDGPS). Our methodology aims to bolster confronted restricted samples. The module integrates frequency-domain random masked patches learning, leveraging feature representations observation distributions improve discriminative properties generated Furthermore, a module, utilizing particle swarm optimization algorithm, refines through pre-trained model. This process reduce errors caused masking. experimental results on four datasets show that proposed model consistently surpasses other methods, achieving highest accuracy F1-score across all datasets: 95.81%, 87.64%, 82.31%, 100% accuracy; 95.21%, 82.32%, 78.57%, F1-Score Epilepsy, FD-B, Gesture, EMG, respectively. In addition, evaluations reinforcement scenario confirm MDGPS’s superior performance. Ablation visualization experiments further validate effectiveness its individual components.