Start date: 01/04/2024
End date: 31/03/2027
ITML Role: Task leader
Budget: 8991728.75€

Summary

The integration of the Internet of Things (IoT) with AI, known as AIoT, generates vast amounts of data critical for improving AI systems and enabling smarter, more efficient operations in vital business and infrastructure areas. However, challenges such as the high costs of data collection, labour intensity, limitations in data availability and scale, and confidentiality rights restrict the availability of realistic datasets. These constraints necessitate the use of offline modelling and simulation, which struggle to accurately represent real-world environments due to variability in human behaviour and events. While Machine Learning (ML) techniques aim to bridge the gap between real and synthetic data, their application is complex without a thorough understanding of the semantics of smart spaces. Real data in usable formats for AIoT systems is often scarce, posing difficulties in training AI models. Moreover, datasets are frequently imbalanced, with common behaviours over-represented, and the dynamic nature of smart spaces requires AIoT systems to adapt continually to changes and unmodeled phenomena.  

To address these challenges, the PANDORA project is developing a framework to prepare and deliver comprehensive datasets for AI models in AIoT systems, enhancing their autonomy, trustworthiness, and energy efficiency. The project objectives include advancing research excellence in resilient, transparent, and human-centered AI approaches for optimized and autonomous data processing and use. It aims to provide novel methods, mechanisms, and tools for creating customizable and trustworthy datasets for model-based AI developments, supporting robust and energy-efficient “data in AI” pipelines across the computing continuum. PANDORA also seeks to deliver a cross-sector and multi-variant smart data space to implement its framework and validate data-enabled trustworthy AI mechanisms in real-life scenarios. Furthermore, it fosters synergetic approaches within EU industrial and scientific research communities, promotes international collaboration on efficient and trustworthy AI methods, and enhances multidisciplinary competencies in industrial AI, data, and robotics while embracing open innovation.

ITML’s role

ITML focuses on data summarization, dimensionality reduction, and fusion. It aims to simplify datasets, reducing redundancy and complexity using Visual Analytical Assessment (VAT) algorithms for unsupervised learning and deep learning for multi-model data summarization. This task enhances data usability by transforming it from raw to semantic levels, crucial for AIoT system training and efficiency. Moreover, ITML ensures continuous data collection and analysis for PANDORA’s experiments. It oversees the technical, legal (including GDPR compliance), and commercial aspects of data sharing, enhancing data assets for trial scenarios and addressing privacy and ethical concerns.