In an era where food safety is paramount, the integration of Artificial Intelligence (AI) offers transformative solutions. The Extreme Food Risk Analytics (EFRA) project, exemplifies this by developing privacy-preserving AI models that provide accurate food risk predictions with enhanced explainability. Recognized by the European Commission’s Innovation Radar, this innovation is poised to revolutionize how we approach food safety.
The EFRA Project: An Overview
EFRA aims to tackle significant scientific, economic, and societal challenges related to the safety and quality of food consumed by Europeans. The project’s objectives include:
1. Data Discovery and Integration: Developing solutions to discover and distill food risk data from diverse and dispersed sources promptly, ensuring timely and relevant information.
2. User-Centric Design: Focusing on human-centric interactions to assess the system’s usefulness for real-world risk prevention actions, ensuring practicality and effectiveness for end-users.
3. Trustworthy AI Systems: Demonstrating the development of trustworthy, accurate, green, and fair AI systems for food risk prevention, ensuring reliability and sustainability.
4. Technological Integration: Integrating relevant technologies such as big data, IoT, and AI to achieve its goals and foster links to respective communities of data innovators in the food supply chain.
To achieve these goals, EFRA is designing, testing, and deploying tools, as well as undertaking initiatives to facilitate their uptake, elicit feedback, and engage stakeholders. The EFRA tools include:
- EFRA Data Hub: Providing a Kubernetes based infrastructure for collecting and integrating food safety datasets from dispersed, multilingual and heterogeneous data sources. It is populated from the marketplace (uploading private datasets from users/clients) and from intelligent crawlers that search, aggregate and link data collected from the web.
- EFRA Analytics Powerhouse: providing a Kubernetes based infrastructure for running AI models and algorithms that distils useful insights & signals from the datasets stored in the EFRA Data Hub.
- EFRA Federated Learning component: a specific component where multiple entities collaboratively train an AI model while keeping their data decentralized and private.
- EFRA Data & Analytics Marketplace: A user-friendly web application that allows interested users to discover, purchase/use, and contribute data, AI models, and analytics modules, creating an economy where data holders and data consumers engage and trade.
Privacy-Preserving AI Models with Enhanced Explainability
A cornerstone of EFRA’s innovation is the development of AI models that not only predict food risks accurately but also preserve data privacy and offer enhanced explainability. In the context of food safety, these models analyze vast amounts of data from various sources to identify potential risks. However, handling such data necessitates stringent privacy measures to protect sensitive information.
EFRA addresses privacy-preserving concerns by adopting a Federated Learning Architecture (based on the well known Flower.AI framework). This technology enables the training of AI models on distributed food safety datasets, ensuring GDPR compliance and maintaining data confidentiality; different data owners can collaborate in the development of AI models without disclosing their own data. This architecture also addresses the challenge of collaboratively learning from non-independently and identically distributed (non-IID) training data. A pioneering AI model has been developed for the outlier detection in poultry mortality rates and end-cycle weights.
Moreover, enhanced explainability ensures that the predictions made by these models are transparent and understandable to stakeholders, fostering trust and facilitating informed decision-making.
Market Maturity and Innovation Radar Recognition
The European Commission’s Innovation Radar has assessed EFRA’s innovation as “Tech Ready”, indicating that it is progressing well in the technology development process, including pilots, prototypes, and demonstrations. This assessment underscores the innovation’s readiness for practical application and its potential impact on the market.
Furthermore, the innovation has been identified as addressing the needs of existing markets and customers, highlighting its relevance and applicability in the current food safety landscape.
The EFRA project stands at the forefront of integrating AI into food safety, offering innovative solutions that are both privacy-preserving and explainable. With recognition from the Innovation Radar and support from Horizon Europe, EFRA is well-positioned to make significant strides in transforming food risk prediction and prevention, ultimately contributing to a safer and more secure food supply chain for all.
