Towards Verified Artificial Intelligence

Abstract

Verified artificial intelligence (AI) is the goal of designing AI-based systems that that have strong, ideally provable, assurances of correctness with respect to mathematically-specified requirements. This paper considers Verified AI from a formal methods perspective. We describe five challenges for achieving Verified AI, and five corresponding principles for addressing these challenges.

References in corpus (5)

• DeepFool: a simple and accurate method to fool deep neural networks
• Constrained Sampling and Counting: Universal Hashing Meets SAT Solving
• Learning Task Specifications from Demonstrations
• Systematic Testing of Convolutional Neural Networks for Autonomous Driving
• Developing Bug-Free Machine Learning Systems With Formal Mathematics

Cited by in corpus (25)

• Algorithmic decision-making in AVs: Understanding ethical and technical concerns for smart cities
• AI Safety Gridworlds
• A Review of Testing Object-Based Environment Perception for Safe Automated Driving
• Software Engineering for AI-Based Systems: A Survey
• Practical Solutions for Machine Learning Safety in Autonomous Vehicles
• Using Machine Learning Safely in Automotive Software: An Assessment and Adaption of Software Process Requirements in ISO 26262
• TEASER: Fast and Certifiable Point Cloud Registration
• If a Human Can See It, So Should Your System: Reliability Requirements for Machine Vision Components
• White Paper Machine Learning in Certified Systems
• A Formalization of Robustness for Deep Neural Networks
• VERIFAI: A Toolkit for the Design and Analysis of Artificial Intelligence-Based Systems
• Real-time Out-of-distribution Detection in Learning-Enabled Cyber-Physical Systems
• Certifiable Artificial Intelligence Through Data Fusion
• Gray-box Adversarial Testing for Control Systems with Machine Learning Component
• Verifier Theory and Unverifiability
• Teaching Temporal Logics to Neural Networks
• On Controllability of AI
• Formal methods and software engineering for DL. Security, safety and productivity for DL systems development
• VeriDL: Integrity Verification of Outsourced Deep Learning Services (Extended Version)
• nn-dependability-kit: Engineering Neural Networks for Safety-Critical Autonomous Driving Systems
• Runtime Interchange for Adaptive Re-use of Intelligent Cyber-Physical System Controllers
• Position Paper: Towards Transparent Machine Learning
• On Testing of Samplers
• Formal Verification of Safety Critical Autonomous Systems via Bayesian Optimization
• Does the dataset meet your expectations? Explaining sample representation in image data