Self-paced data augmentation for training neural networks

Data augmentation is widely used for machine learning; however, an effective method to apply data has not been established even though it includes several factors that should be tuned carefully. One such factor sample suitability, which involves selecting samples are suitable augmentation. A typical applies all training disregards may reduce classifier performance. To address this problem, we propose the self-paced (SPA) automatically and dynamically select when a neural network. The proposed mitigates deterioration of generalization performance caused by ineffective We discuss two reasons SPA works relative curriculum learning desirable changes loss function instability. Experimental results demonstrate can improve performance, particularly number small. In addition, outperforms state-of-the-art RandAugment method.