Can we Defend Against the Unknown? An Empirical Study About Threshold Selection for Neural Network Monitoring
This addresses a critical gap in runtime monitoring for neural networks in safety-critical systems, but it is incremental as it builds on existing monitoring techniques.
The study tackles the problem of selecting thresholds for neural network monitors to reject unsafe predictions, finding that current methods often fail against unforeseen threats and that integrating generic threats can improve robustness.
With the increasing use of neural networks in critical systems, runtime monitoring becomes essential to reject unsafe predictions during inference. Various techniques have emerged to establish rejection scores that maximize the separability between the distributions of safe and unsafe predictions. The efficacy of these approaches is mostly evaluated using threshold-agnostic metrics, such as the area under the receiver operating characteristic curve. However, in real-world applications, an effective monitor also requires identifying a good threshold to transform these scores into meaningful binary decisions. Despite the pivotal importance of threshold optimization, this problem has received little attention. A few studies touch upon this question, but they typically assume that the runtime data distribution mirrors the training distribution, which is a strong assumption as monitors are supposed to safeguard a system against potentially unforeseen threats. In this work, we present rigorous experiments on various image datasets to investigate: 1. The effectiveness of monitors in handling unforeseen threats, which are not available during threshold adjustments. 2. Whether integrating generic threats into the threshold optimization scheme can enhance the robustness of monitors.