Rethinking RobustBench: Is High Synthetic-Test Data Similarity an Implicit Information Advantage Inflating Robustness Scores?

Computer Science

• Implicit Information 100%
• Synthetic Datasets 100%
• Training Data 100%
• Similarity Data 100%
• Large Data Set 50%
• Feature Space 50%
• Performance Gain 50%
• Massive Datasets 50%
• Similarity Score 50%
• Level Similarity 50%
• Adversarial Robustness 50%