Descripción del proyecto
Abordar la avalancha de datos con técnicas de vanguardia
La actual proliferación de aplicaciones intensivas en datos sobrecarga la infraestructura existente, lo cual provoca ineficiencias y una gestión de recursos subóptima. Los métodos tradicionales tienen dificultades para afrontar la naturaleza dinámica de las demandas informáticas modernas, especialmente en entornos periféricos y en la nube. Dicha fragmentación dificulta la escalabilidad, elasticidad y portabilidad, obstaculizando el funcionamiento fluido de las aplicaciones distribuidas. En este contexto, el equipo del proyecto ENACT, financiado con fondos europeos, desarrollará un continuo de computación cognitiva. Asimismo, aprovecha modelos gráficos dinámicos para visualizar el estado de los recursos en tiempo real, ayudando a modelos de inteligencia artificial como las redes neuronales gráficas y los agentes de aprendizaje por refuerzo profundo a sugerir configuraciones de despliegue óptimas. Tales avances allanan el camino a motores inteligentes de toma de decisiones, revolucionando la gestión de infraestructuras. El equipo de ENACT también es pionero en un modelo de programación de aplicaciones que admite aplicaciones autodeterminadas para diversos recursos.
Objetivo
ENACT develops cutting-edge techniques and technology solutions to realise a Cognitive Computing Continuum (CCC) that can address the needs for optimal (edge and Cloud) resource management and dynamic scaling, elasticity, and portability of hyper-distributed data-intensive applications. At infrastructure level, the project brings visibility to distributed edge and Cloud resources by developing Dynamic Graph Models capable of capturing and visualising the real-time and historic status information, connectivity types, dependencies, energy consumption etc. from diverse edge and Cloud resources. The graph models are used by AI (Graph Neural Networks - GNN) models and Deep Reinforcement Learning (DRL) agents to suggest the optimal deployment configurations for hyper distributed applications considering their specific needs. The AI (GNN and DRL) models are packaged as an intelligent decision-making engine that can replace the scheduling component of open-source solutions such as KubeEdge. This will enable real-time and predictive management of distributed infrastructure and applications. To take full advantage of the potential (compute, storage, energy efficiency etc) opportunities in the CCC, ENACT will develop an innovative Application Programming Model (APM). The APM will support the development of distributed platform agnostic applications, capable of self-determining their optimal deployment and optimal execution configurations while taking advantage of diverse resources in the CCC. An SDK to develop APM-based distributed applications will be developed. Moreover, services for automatic (zero-touch provisioning-based) resource configuration and (telemetry) data collections are developed to help design and update dynamic graph models. ENACT CCC solutions will be validated in 3 use-cases with challenging resource and application requirements. International collaboration is planned as Japan Productivity Center has committed to support with knowledge sharing.
Ámbito científico
- natural sciencescomputer and information sciencesartificial intelligencemachine learningreinforcement learning
- social scienceseconomics and businesseconomicsproduction economicsproductivity
- natural sciencesmathematicspure mathematicsdiscrete mathematicsgraph theory
- natural sciencescomputer and information sciencesartificial intelligencecomputational intelligence
Palabras clave
Programa(s)
Régimen de financiación
HORIZON-RIA - HORIZON Research and Innovation ActionsCoordinador
57001 Thermi Thessaloniki
Grecia