ROAST: Risk-aware Outlier-exposure for Adversarial Selective Training of Anomaly Detectors Against Evasion Attacks

Dimensions overall score 4.0

GitHub Code Pulse

No public code linked for this paper yet.

• ROAST improves anomaly detector recall by 16.2% on average compared to indiscriminate training.result 95%
• ROAST reduces training time by 88.3% on average compared to indiscriminate training.result 95%
• ROAST focuses training on data from less vulnerable patients to improve anomaly detector recall.method 90%
• ROAST applies outlier exposure by injecting adversarial samples into the training set of less vulnerable patients to preserve precision.method 90%
• ROAST achieves a recall improvement of 16.2% on average with a precision reduction in only one out of nine experiments.result 95%
• ROAST requires significantly fewer training samples on specific medical datasets compared to indiscriminate training.result 90%
• Conventional anomaly detectors trained indiscriminately on all patient data overlook physiological differences, leading to noise and degraded robustness.limitation 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card