LATTE: <u>L</u> STM Self- <u>Att</u> ention based Anomaly Detection in <u>E</u> mbedded Automotive Platforms

Modern vehicles can be thought of as complex distributed embedded systems that run a variety automotive applications with real-time constraints. Recent advances in the industry towards greater autonomy are driving to increasingly connected various external (e.g., roadside beacons, other vehicles), which makes emerging highly vulnerable cyber-attacks. Additionally, increased complexity and in-vehicle networks results poor attack visibility, detecting such attacks particularly challenging systems. In this work, we present novel anomaly detection framework called LATTE detect cyber-attacks Controller Area Network (CAN) based within platforms. Our proposed uses stacked Long Short Term Memory (LSTM) predictor network attention mechanisms learn normal operating behavior at design time. Subsequently, scheme (also trained time) is used (as anomalies) runtime. We evaluate our under different scenarios detailed comparison best-known prior works area, demonstrate potential approach.