An Incremental Learner for Language-Based Anomaly Detection in XML

Explore an innovative approach to XML security through an 18-minute IEEE conference talk on incremental learning for language-based anomaly detection. Delve into the challenges posed by extension points in XML Schema validation and discover a novel solution using datatyped XML visibly pushdown automata. Learn how this incremental learner infers types and datatypes in mixed-content XML, creating a representation free from extension points and capable of stream validation. Examine the learner's convergence guarantees, unlearning and sanitization operations, and its practical applications in web service scenarios. Gain insights into how this method outperforms traditional schema validation, achieving zero false positives in simulated XML attacks.