Sketching Curvature for Efficient Out-of-Distribution Detection for Deep Neural Networks

Keyphrases

• Deep Neural Network 100%
• Out-of-distribution Detection 100%
• Training Distribution 75%
• OOD Detection 75%
• Training Data 50%
• Uncertainty Estimation 50%
• Distribution Detection 50%
• Epistemic Uncertainty 50%
• Local Curvature 25%
• Detection Performance 25%
• Decision System 25%
• Task-irrelevant 25%
• Training Model 25%
• Low Computational Complexity 25%
• Agnostic 25%
• Real-time Decision Making 25%
• Test Input 25%
• Fisher Information Matrix 25%
• Data Sketches 25%
• Low-rank Approximation 25%
• Weighted Spaces 25%
• Matrix Sketching 25%
• Pre-trained Networks 25%

Computer Science

• Deep Neural Network 100%
• Training Data 50%
• Decision-Making 25%
• Detection Performance 25%
• Input Distribution 25%
• Local Curvature 25%
• Rank Approximation 25%
• Fisher Information Matrix 25%
• Trained Network 25%

Mathematics

• Deep Neural Network 100%
• Training Data 50%
• Epistemic Uncertainty 50%
• Matrix (Mathematics) 25%
• Low-Rank Approximation 25%
• Fisher Information Matrix 25%
• Weight Space 25%