Pure Noise to the Rescue of Insufficient Data: Improving Imbalanced Classification by Training on Random Noise Images

Keyphrases

• Low Quality Data 100%
• Noisy Image 100%
• Random Noise 100%
• Imbalanced Classification 100%
• Training Data 50%
• Highly Effective 25%
• Deep Neural Network 25%
• Additive Noise 25%
• Effective Method 25%
• Image Classification 25%
• Overfitting 25%
• Training Procedure 25%
• Long Tail 25%
• Classification Performance 25%
• Visual Recognition 25%
• Recognition Task 25%
• Additional Training 25%
• Data Augmentation 25%
• Generation Method 25%
• Natural Images 25%
• Real-world Examples 25%
• Adversarial Noise 25%
• New Distribution 25%
• Classification Datasets 25%
• ImageNet 25%
• CIFAR-100 25%
• Training Scheme 25%
• Batch Normalization Layer 25%
• CelebA 25%

Computer Science

• Image Noise 100%
• Training Data 50%
• Image Classification 25%
• Classification Performance 25%
• Effective Method 25%
• Data Generation 25%
• Data Augmentation 25%
• Batch Normalization 25%