Towards Assurance-Based Learning-Enabled Cyber-Physical Systems
Institute for Pure & Applied Mathematics (IPAM) via YouTube
Overview
Syllabus
Intro
DARPA - Assured Autonomy Program Goal
Illustrating the challenge
Safety Assurance for Systems - State of Practice
VU ALC Project vision
Project activities
AUV as a CPS with LECS Learning-Erabled Components
LEC Verification: Reachability Analysis of Feedforward/Convolutional Neural Networks
Closed-Loop CPS with LECS: Verification Flow and Tools
UUV Closed-Loop Verification Results
ACAS-Xu Closed-Loop Scenario
ACAS-Xu Closed-Loop Verification Results
Perception Robustness Verification Target application Perception LE Component
Semantic Segmentation Robustness
Assurance Monitoring
Inductive Conformal Prediction (ICP)
Anomaly Detection
Exchangeability Martingales
Distribution Shift Detection Example
Distribution Shift Detection in adversarial scenarios
Tool architecture coverage
ALC Toolchain Our approach
ALC Design Workflow Specialized for LEC development
Modeling Blocks, Systems, Training, & Execution
System Modeling
System architecture SysML block diagrams
Data Collection
Testing
Automation: Workflow Models
Toolchain Support for Data Provenance
System-level Assurance Technology Dynamic assurance
Taught by
Institute for Pure & Applied Mathematics (IPAM)