Events

AI systems need constant maintenance to provide accurate and precise results post-deployment. In critical infrastructure systems—where a system might be air gapped or completely segmented from the rest of the systems—questions arise of when do I know when my model needs to be updated? And how can I ensure my model is accurate and precise long-term? We present a method of quantifying resilience in post-deployment conditions using statistical methods and discuss how this method can be applied to determining when a model needs to be updated, including how long a model is accurate and precise, and how adversarial action influences resilience.