AI for Design and Verification of Safety/Mission Critical Intelligent Systems
Activity (Attività Formativa Complementare, AFC) offered within the Master Programme in Computer Science at Sapienza University of Rome.
Intelligent Systems (ISs) are systems with the capability to sense the environment and react to it to attain given goals. Often, intelligent systems have two main components: one focusing on learning the environment behaviour on the basis of the acquired measurements (typically based on data-driven Artificial Intelligence methods, namely Machine Learning) and another one focusing on planning the best course of actions given what has been learned so far (typically relying on model-based Artificial Intelligence approaches).
Examples of intelligent systems are: software controlling autonomous vehicles (e.g., cars, drones, swarm of drones, satellites, vessels, robots, etc.); software controlling autonomous systems in general (e.g., bio-medical devices, trading-robots, automatic video compression, decision support systems, etc.).
Intelligent systems are often mission-critical (i.e., their failure leads to a loss of money, e.g., a trading-robot, a satellite, a swarm of drones) or safety-critical (i.e., their failure may lead to a loss of human lives, e.g., bio-medical devices, autonomous driving).
Designing and verifying intelligent systems is very challenging because of the wide set of operational scenarios (environment behaviours) they are supposed to withstand and the fact that the system behaviours change (through learning) as a function of the inputs received.
Involved students will see that design and verification of an intelligent system can be both cast as (quite large) optimisation problems and solved using a suitable blend of Artificial Intelligence approaches, both model-based (e.g., knowledge representation and reasoning) and data-driven (Machine Learning, ML).
The activity has the following goals:
- Learn how to use hybrid Artificial Intelligence techniques to automatically build an environment model from historical (sensor) data (statistics and ML) and background knowledge (model-based AI). Such a model will be used as an adversarial scenario generator to support design and verification of ISs.
- Learn how to use hybrid AI techniques (both model-based and data-driven) to develop suitable Key Performance Indicators (KPIs) defining functional as well as non-functional requirements for ISs.
- Learn how to use AI (both model-based and data-driven) and Statistical Model Checking techniques to support simulation-based design and verification of ISs.
- Apply the above techniques to a specific domain, e.g.: Healthcare, Cyber-Physical Systems, Energy, Trading-Robots.