BadLabel: A Robust Perspective on Evaluating and Enhancing Label-Noise Learning

Keyphrases

• Noisy Label Learning 100%
• Learning Algorithm 50%
• Noise Type 37%
• Noisy Labels 37%
• Label Noise 37%
• Learning Methods 25%
• Value Loss 25%
• Clean Label 25%
• Source Code 12%
• Small Sets 12%
• Labeled Data 12%
• Semi-supervised Learning 12%
• Training Data 12%
• Learning Objectives 12%
• Generalization Performance 12%
• New Dataset 12%
• Novel Labels 12%
• Large Margin 12%
• Model Generalization 12%
• Instance-dependent Noise 12%
• Label Flipping Attack 12%
• Data with Imperfect Labels 12%

Computer Science

• Learning Algorithm 100%
• Existing Label 50%
• Experimental Result 25%
• Source Codes 25%
• Training Data 25%
• Semisupervised Learning 25%
• Generalization Performance 25%

Physics

• Noise Type 100%
• Source Coding 33%
• Supervised Learning 33%