New efficient, autonomous and self-configurable Artificial Intelligence techniques and models for sustainable data

The ORLEG-IA project will develop and validate new sustainable and efficient machine learning methods by design. Its sustainability will be approached with a holistic perspective, from sensor-based data acquisition (Green Sampling) to incremental and continuous learning (Green Continuous Learning) and automatic model design with efficiency criteria (Green Model Design). As a scope of application, the methods developed will be tested in continuous monitoring scenarios of biological signals for the detection of pathologies (e.g., arrhythmias on an electrocardiogram) or the characterization of the cognitive state of patients using cognitive status of patients with motion sensors.

Applicability/impact of results:
From an industrial point of view, ORLEG-IA will enhance the scientific-technological positioning of its member institutions, and the competitiveness of Basque companies in different sectors, making it possible to:

• Reducing the development time of AI-based solutions by data scientists through the automation of many tasks (data cleaning and organisation, validation and monitoring of information, model validation and monitoring, attribute selection) The use of the model is not only a matter of the model's own development, but also a matter of the model's own development.
• Simplify the validation of models, facilitating the adaptation of their configuration to the conditions of uncontrolled environments, where data and modelling tasks may change over time.
• Reduce operational risks associated with data errors, implementation, inappropriate use or adaptation to changing conditions under which models have not been designed or trained.
• Advance model quality, traceability and comparability through automation of model maintenance.
• Improve the energy efficiency of models by automating the maintenance of the algorithms.
• Improve the energy efficiency of models by enabling their training and execution on low-cost, resource-constrained devices (wearables, mobile devices), which require less power consumption associated with the AI models that use their applications.
• Minimise reliance on highly skilled AI/ML staff in the deployment of AI-based solutions.

At the academic level, based on the algorithms and results achieved in this project, we will seek collaborations with the University of Bordeaux, which is part of the Bordeaux Alliance for Artificial Intelligence, along with other industrial players.