Responsable: Rocío González Díaz
Tipo de Proyecto/Ayuda: Proyectos de Transición Ecológica y Transición Digital
Referencia: TED2021-129438B-I00
Fecha de Inicio: 01-12-2022
Fecha de Finalización: 30-11-2024

Empresa/Organismo financiador/es:

• Ministerio de Ciencia e Innovación

Equipo:

• Equipo de Investigación:
  • Miguel Ángel Gutiérrez Naranjo
  • María José Jiménez Rodríguez
  • Eduardo Paluzo Hidalgo (alta: 23/01/2024)
• Equipo de Trabajo:
  • José Luis Fernández Rueda
  • Manuel Soriano Trigueros
  • Aurelio Barrera Vicent (alta: 08/02/2023)
  • Javier Perera Lago (alta: 04/06/2023)
  • Víctor Toscano Durán (alta: 08/02/2023)

Contratados:

• Investigadores:
  • Álvaro Torras Casas

Resumen del proyecto:

Artificial intelligence (AI) is being widely applied in many domains at a very fast pace. Its development has significantly accelerated in the last decade due to different concurrent technological advances, e.g., the development of increasingly powerful computational architectures and algorithms, novel hardware architectures, and the availability of large volumes of data. In the last few years, these architectures have achieved unprecedented performance in different fields. However, despite this success, the adverse environmental impacts of AI could jeopardize the achievement of the goals of the European Green Deal. Recent studies demonstrate that the computations required to train AI models have a surprisingly large carbon footprint. Implementing efficiency in AI research will reduce its carbon footprint and increase its inclusion. At present, it is difficult to find an area of AI whose state-of-the-art is not dominated by AI solutions, even in problems where AI models did not seem to be applicable, such as symbolic computations or language translation. AI solutions are developed and oriented to work on huge datasets, both dimensionally and in the number of samples.
In this proposal, we aim to provide topology-based methodology towards a green AI, pursuing an ecological transition and sustainable implementations of these models. We can summarize the contribution of this project as follows: 1) we plan to study dataset size reduction in terms of size and dimensionality from a computational topology perspective, towards efficiency and energy saving starting from our seminal work on representative datasets1; 2) we plan to develop new AI models and improve current ones towards topology-aware trustworthiness and sustainable implementations such as simplicial-map neural networks23. Finally, we will ensure robustness through mathematical theoretical foundations. All this follows open science and open data philosophy.