PhD Project 1: Regulation-compliant test case and oracle generation for AI-enabled CPSs

  • Host Name: Mondragon University - University Web Page
  • Main Supervisor: Aitor Arrieta - Supervisor Web Page
  • Project Objectives: The goal of this DC is to develop an automated testing framework for AI-enabled CPSs that should comply with different AI regulations. This DC will focus on test input and test oracle generation, that is, it will generate cost-effective test cases that aim at testing AI-enabled CPSs in compliance with different regulations in an automated manner. To this end, the DC, together with DC7, will first study the EU-AI Act regulation to see which aspects are relevant for AI-enabled CPSs. It will complement this with other sector-specific regulations (e.g., Regulation (EU) 2018/858), to see the commonalities and variabilities of the regulations in relation to different types of CPSs. Based on this first investigation, the DC will select those specific aspects from the regulations relevant to AI-enabled CPSs that can be automated within the framework. The DC will implement a tool that uses, among others, Large Language Models (LLMs) to generate test inputs and oracles for the CPS and checks its compliance with relevant laws and regulations automatically. Test cases will be both for off-line testing (i.e., only the AI component) as well as on-line testing (the AI components with the physical part of the CPS).
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 2: Automated Program Repair of ACPSs

  • Host Name: Mondragon University - University Web Page
  • Main Supervisor: Aitor Arrieta - Supervisor Web Page
  • Project Objectives: The goal of this DC is to develop an automated program repair (APR) framework for ACPSs. While APR has been a widely investigated area in the context of software systems, its application to CPSs face several scalability issues due to two core challenges: (1) the use of simulation-based testing, and as a direct consequence, the test execution time; and (2) the use of a low number of test cases to check whether a patch is valid or not. This DC will have as a goal to solve these challenges by means of three sub-goals. The first sub-goal will be to develop a method to localize the bug, which can either be at the AI component (i.e., neural network) or at the code level. The second sub-goal will be to develop a novel method to repair code-level bugs, which will combine search-based techniques with LLMs to propose patches. The last sub-goal will be to develop a novel method to repair neural-network level bugs, which can be solved by taking actions like re-training by amplifying the training dataset, changing the network architecture or changing training hyperparameters. To validate the changes, it will use test cases and oracles from the methods developed by DC1 as well as the tools generated by DC6.
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 3: HMI-based run-time monitoring for AI-enabled CPSs.

  • Host Name: Mondragon University - University Web Page
  • Main Supervisor: Maitane Mazmela - Supervisor Web Page
  • Project Objectives: Under the EU-AI Act proposal, high-risk AI systems shall be designed and developed in such a way, including with appropriate human-machine interface (HMI) tools, that they can be effectively overseen by natural persons during the period in which the AI system is in use. To this end, effective and efficient run-time monitoring approaches are necessary. AI-enabled CPSs should be safe even under unforeseen and uncertain situations. The goal of this DC will be, on the one hand, to research effective and efficient run-time monitoring approaches for AI-enabled CPSs, in collaboration with DCs 4, 13 and 14. And, on the other hand, to research effective ways for communicating this to the human. For instance, in an autonomous vehicle, when an unforeseen situation is given and the system is in an unsafe situation, this should be effectively indicated to the human driver, so as to take the appropriate action. With this, a framework to design HMI for ACPSs will be developed and validated in an open-source case study and the ITDP.
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 4: Developing and evolving digital twins with model-based and data-driven learning approaches for uncertainty handling in operation

  • Host Name: Oslo Metropolitan University - University Web Page
  • Main Supervisor: Shaukat Ali - Supervisor Web Page
  • Project Objectives: 1) Develop a digital twin engineering methodology including hardware, physical environment, software, and network modeling with model-based systems engineering (MBSE) approaches for co-simulation, and evolve these models by continuously learning digital twin models from data (historical and real-time data) with relevant machine learning techniques; 2) Develop novel holistic uncertainty quantification methods for deep learning models embedded in ACPS, and ACPSs as a whole; 3) Develop uncertainty analyses and handling methods (i.e., anomaly detection, robustness assessment, probabilistic uncertainty forecasting) relying on machine learning techniques, search algorithms, and relevant theories (e.g., probability and uncertainty) to enable uncertainty handling of operational ACPSs via their digital twins; 4) Devising novel methods to autonomously handling unsafe situations when those are detected.
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 5: Uncertainty quantification and handling of large language models for ACPS quality engineering tasks

  • Host Name: Simula Research Laboratory - Research Center Web Page
  • Main Supervisor: Shaukat Ali - Supervisor Web Page
  • Project Objectives: 1) Investigating existing metrics to quantify and estimate uncertainty in LLMs such as single-inference and multi-inference and assessing their strengths and weakness in our context; 2) Devising novel metrics for uncertainty quantification and estimated based on identified weaknesses in the first objective; 3) Devising novel methods to use quantified uncertainty for risk assessment of the use of LLMs for ACPS quality engineering tasks to assess their trustworthiness, 4) Studying relationships of uncertainty with correctness and other non-functional characteristics (e.g., performance) of quantum engineering artifacts produced by LLMs such as test cases
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 6: Testing ACPSs with hybrid methods

  • Host Name: Simula Research Laboratory - Research Center Web Page
  • Main Supervisor: Shaukat Ali - Supervisor Web Page
  • Project Objectives: 1) Develop novel methods to test ACPS with a combination of AI methods such as reinforcement learning and evolutionary algorithms in software in the loop simulation, including generating multi-modal environmental scenarios and uncertain situations; 2) Combining previous AI-based methods with LLMs to iteratively improve the realism of scenarios and uncertain situations including efficient prompt engineering, and incorporating risk assessment and uncertainty quantification methods from DC5 to improve the quality and realism of scenario generation further, while at the same time reducing the uncertainty in LLMs, 3) Releasing the integrated testing framework as open source and extensively assessing its cost and effectiveness with various ACPSs and publicly available LLMs.
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 7: A check-list generation framework for AI-enabled CPS developers to support compliance of the system with regulations

  • Host Name: University of Malaga - University Web Page
  • Main Supervisor: Pablo Sanchez - Supervisor Web Page
  • Project Objectives: The goal is to develop a tool that will support CPS developers to comply with different legislative regulations by the EU and associated countries when developing their systems. To this end, the DC will first study the EU-AI Act regulation to see which aspects are relevant to AI-enabled CPSs, with the help of DC1. It will also complement this with other sectors specific to different CPSs. From there, it will extract commonalities and variabilities of the law based on the systems they are targeting to develop, which will be the base for developing the tool. Afterwards, the DC will develop a software tool for CPS developers. The tool will include an interface where the user will introduce the different characteristics of the system. The back-end of the tool will process all the data and will provide as an output a document that will provide a set of checklist for the user to help them build a system compliant with the EU regulation.
  • Required Qualifications: Bacherlor degree and master's study in law.

PhD Project 8: Analyzing and Improving the legal protection of AI

  • Host Name: University of Malaga - University Web Page
  • Main Supervisor: Pablo Sanchez - Supervisor Web Page
  • Project Objectives: The objective is to analyze the regulations on AI in the EU and in those Member States with more advanced regulation in order to ascertain the current state of affairs. A preliminary analysis allows us to affirm the incipient state of these regulations and the need to implement improvements. As AI makes autonomous decisions, the question arises as to who is liable in case of harm or damage caused by an AI system. New rights related to legal responsibility and accountability of AI developers and systems may need to be established. Therefore, this DC aims to develop novel regulatory frameworks that can be applied in the context of dependable AI-intensive systems (including ACPS). The proposed framework will be generic enough to be applicable in different countries (especially within the EU), and able to fulfill the current EU regulations. Also, upon developing the framework, the DC will also develop technical requirements that impact development activities (e.g., analysis and testing) and artifacts (e.g., Software and AI Bills of Materials).
  • Required Qualifications: Bacherlor degree and master's study in law.

PhD Project 9: Studying, identifying, and fixing technical debt in LLM-intensive systems

  • Host Name: University of Sannio - University Web Page
  • Main Supervisor: Massimiliano Di Penta - Supervisor Web Page
  • Project Objectives: The overall goal of this DC is to study LLM-specific technical debt, and to develop approaches and tools to identify and repair it. 1) Providing a definition and taxonomy of technical debt for LLM-intensive systems. The taxonomy will be empirically elicited by analyzing existing systems, combining quantitative and qualitative methods; 2) Developing an approach aiding to automatically identifying technical debt in LLM-intensive systems. The approach will itself leverage state-of-the-art machine learning techniques, including LLMs, and mine different kinds of artifacts, including source code, models’ metadata, and information originating from MLOps pipelines; 3) Developing approaches repairing technical debt in LLM-intensive guidelines.
  • Required Qualifications: Bach. degree and master's degree in computer science, computer engineering, or related study.

PhD Project 10: Laws and regulations in LLM-based development

  • Host Name: University of Sannio - University Web Page
  • Main Supervisor: Massimiliano Di Penta - Supervisor Web Page
  • Project Objectives: The overall goal of this DC is to study legal problems related to the use of AI-generated code in software systems, and to propose techniques to cope with these problems. The specific sub-objectives are: 1) Identify Intellectual property-related problems related to the use of LLMs in software development. This will be done by conducting surveys, interviews, and by analyzing data from software repositories. 2) Develop an approach for provenance analysis of LLM-recommended source code. The proposed approach would on the one hand leverage existing origin-analysis approaches and large repositories (such as World of Code); 3) Development of an approach for the automated generation of Software Bills of Material for software systems leveraging automatically-generated code. The approach will encompass analysis of hardware configurations for CPSs (to generate HBOMs) and model descriptions or model cards for the generation of AIBOMs.
  • Required Qualifications: Bach. degree and master's degree in computer science, computer engineering, or related study.

PhD Project 11: ACPS Sustainability-Quality Modelling and Planning

  • Host Name: Vrije Universiteit Amsterdam - University Web Page
  • Main Supervisor: Patricia Lago - Supervisor Web Page
  • Project Objectives: The project aims to define the sustainability priorities of ACPSs as quality properties, and to synthesise Decision Maps and related Sustainability-Quality properties and associated metrics for both direct and indirect sustainability impact that should be fed to LLMs or vice versa, be generated by LLMs. To this end, the DC will (1) review existing works addressing quality properties of ACPSs; (2a) propose an operational definition of the extracted suite of quality properties and metrics to quantify, e.g., uncertainty or energy efficiency, based on internal|external sustainability, the well-known sustainability dimensions (technical, social, environmental, economic) and the impact over time; (2b) specialise/extend it for the ACPS domains tackled by other partners; and (3) propose a method to represent measures collected at operation time (WP2) and link them to the ACPS artefacts (e.g., architecture components or architectural tactics) that are responsible for the measured quality so that (4) good software practices (or anti-patterns) can be extracted and made reusable at design time (WP1)
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 12: Conversations with the Architect – Large Language Models for Designing Energy-Efficient ACPSs

  • Host Name: Vrije Universiteit Amsterdam - University Web Page
  • Main Supervisor: Ivano Malavolta - Supervisor Web Page
  • Project Objectives: The project aims to exploit LLMs to better support ACPS developers in developing energy-efficient software for ACPSs. At the core of the project lies the concept of architectural tactic, i.e., design decisions that influence the achievement of system qualities and can be reused across projects. For example, a tactic for energy efficiency is to offload computationally-expensive algorithms from a battery-powered robot to the Cloud. Tactics have been studied and successfully used in areas like big-data cybersecurity and Cloud-based systems, but they have never been used in conjunction with LLMs. LLMs will be used for recommending architectural tactics for ACPSs either conversationally to developers at development time (WP1) or programmatically to the ACPS itself at runtime (WP2). Architectural tactics in ACPSs are highly domain- and context-dependent, they can have side effects, and can come with non- trivial complex trade-offs. The ability to process large amounts of data and internalise implicit cross-domain knowledge of LLMs makes them excellent candidates for managing architectural tactics. The objectives of the project include: (1) to build a knowledge base of concrete, repeatable, and quantifiable architectural tactics for energy-efficient ACPSs, (2) to integrate such knowledge base into different LLMs for providing timely recommendations about applicable tactics at development time, (3) to develop an approach for automatically applying and composing architectural tactics in the context of ACPSs, and (4) to develop a self-adaptive approach where ACPSs autonomously apply tactics to their software architecture in response to changes in their measured quality of service (e.g., energy degradation).
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.

PhD Project 13: Monitoring and self-adaptability to ACPS states

  • Host Name: University of Bern - University Web Page
  • Main Supervisor: Sebastiano Panichella - Supervisor Web Page
  • Project Objectives: A main goal targeted by the project is to enable ACPS to self-adapt to unprecedented scenarios, to fulfill the required level of trustworthiness. To this aim we target to (1) develop facilities leveraging AI methods, meta-heuristics, and LLMs that are able to adaptively learn/detect misbehaviors and unsafe states in ACPS at runtime; and (2) extend the facilities above with the ability to support CPS self-adaptability, e.g., to new scenarios in X-in-the-loop activities (HiL, simulation, etc.) activities
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study

PhD Project 14: Monitoring and self-healing of ACPS quality aspects based on hybrid and generative methods

  • Host Name: University of Bern - University Web Page
  • Main Supervisor: Sebastiano Panichella - Supervisor Web Page
  • Project Objectives: Our objective is to develop a component for assessing the quality aspects of ACPS. We focus on addressing mitigation strategies using hybrid methods (AI, meta-heuristic, LLMs) for: 1) predicting and fixing Flaky Scenarios with X-in-the-loop Facilities, 2) enhancing ACPS quality under security attacks, and 3) minimizing resource usage in X-in-the-loop based on Historical Analysis.
  • Required Qualifications: Bachelor degree and master's degree in computer science or related study.