Session brief
Learned Execution Graphs for Anomaly Detection & Drift in APIs — Ritvik Pandya, JP Morgan Chase
Ritvik Pandya , JP Morgan Chase
Overview
This talk introduces Learned Execution Graphs (LEGs) as a novel approach to detecting anomalies and drift in API usage. The core thesis is that by modeling API interactions as graphs and learning their execution patterns, systems can proactively identify deviations from normal behavior, which is crucial for maintaining API reliability and security.
Who should watch
- AI engineers working with APIs
- Developers responsible for API reliability and monitoring
- Security professionals concerned with API abuse
- Anyone building or maintaining complex API-driven systems
- Teams experiencing unexpected API behavior or performance issues
Key takeaways
- Learned Execution Graphs (LEGs) represent API calls as nodes and sequences as edges, allowing for the learning of normal interaction patterns.
- LEGs can detect anomalies by identifying deviations from learned graph structures, such as unexpected sequences of calls or calls to non-existent endpoints.
- The system models API drift by tracking changes in the execution graph over time, enabling proactive adjustments or alerts.
- This approach moves beyond simple rate limiting or static rule-based detection to a more dynamic and context-aware anomaly detection method.
- LEGs can be trained on historical API usage data to establish a baseline of expected behavior.
- The framework aims to improve API security and stability by catching subtle or novel forms of misuse.
Notable quotes
Ritvik Pandya discussed modeling API interactions as graphs to learn normal execution patterns.
The system identifies anomalies by detecting deviations from these learned graph structures.
Unofficial community note. Prefer the recording for nuance.