New projects aim at enabling the practical adoption of trustworthy AI
Three new projects at the CAI address key challenges for practical adoption of reliable AI, comprising quality control, development, testing, and certification.
Three newly awarded projects reinforce the mission of the ZHAW Center for Artificial Intelligence (CAI) to enable the practical adoption of Artificial Intelligence (AI). These projects address key aspects of safe and reliable AI deployment comprising quality control, development and testing and certification of AI systems.
In the first project, termed DISTRAL (“Industrial Process Monitoring for Injection Molding with Distributed Transfer Learning”) and funded by Innosuisse, ZHAW CAI partners with Kistler Group and ZHAW Institute of Embedded Systems (InES) to develop a distributed machine learning system to sort out defect plastic parts during production. This project will develop a solution based on distributed machine learning that saves costs, improves usability, and improves production quality. The project entails specific research in transfer learning and federated learning: Using a novel data-centric development process for deep neural networks, it will achieve a semantic transfer of process knowledge that goes far beyond the current state of the art. The resulting model will be able to run on edge devices as well as in the cloud.
The second project, LINA ("Shared Large-scale Infrastructure for the Development and Safe Testing of Autonomous Systems"), will build the largest European infrastructure for research, development, and safe testing of autonomous systems such as drones or service robots. This infrastructure, to be established in the Kanton of Zurich, will comprise a large-scale indoor flight-testing arena, an outdoor physical cage, as well as an outdoor digital cage. Together, these infrastructures will cater to the needs of different stakeholders in the autonomous systems space, providing facilities for research, development and testing from technology readiness levels (TRL) 1, observing basic principles, to TRL 9, prove of actual system in operational environment. This project, funded by the DIZH Innovation program, will be developed in collaboration with the University of Zurich, the ZHAW Center for Aviation (ZAV), and the Zurich University of the Arts (ZHdK) with the support of more than 35 practice partners.
Finally, upcoming regulations will require certain types of AI systems to be certified. However, certification bodies currently lack means that allow them to evaluate all aspects of an AI system, including dimensions such as autonomy and control, transparency, reliability, and safety. In the Innosuisse-funded certAInty project, the CAI addresses this gap by developing a comprehensive framework for evaluation of AI systems comprising the processes for its development and operation (e.g., document management, change management process), as well as the requirements, technical criteria, measures, and actions directly related to the product. As major innovation, our framework will include technical methods, developed in this project, for verifying relevant properties of the system (such as data management, model validation, verification and explainability). Our partners in this project are the ZHAW Institute of Applied Mathematics and Physics (IAMP) and CertX, the first Swiss certification body for functional safety and cyber security of industrial systems. Outcomes of this project will strengthen the recently launched AI certification program (CertAI) launched by CertX, Fraunhofer IAS and MunichRE.
Through these endeavours, CAI addresses key challenges in the development and deployment of reliable, trustworthy AI-powered systems. It leverages its expertise in applied science to advance the state-of-the-art in AI by developing novel methods and infrastructure for improving the reliability of industrial processes and for the certification and validation of AI systems. Altogether, these projects will benefit CAI partners and society by enabling and accelerating the deployment of more efficient, trustworthy products and systems.