A comprehensive review on green AI methods, strategies & measurement frameworks

Résumé

Abstract As Artificial Intelligence (AI) systems continue to scale in complexity and deployment, their environmental impact has become a critical concern. Green AI represents a growing movement toward environmentally conscious model design, advocating for energy-aware practices throughout the development, training, and deployment of AI systems. This paper presents a comprehensive survey of Green AI, aiming to raise awareness of its importance and provide actionable strategies for researchers and practitioners. We begin by outlining the foundational principles and motivations behind Green AI, followed by a review of recent advancements in model compression, reuse, and training optimization techniques. This survey clarifies common misconceptions surrounding Green AI and its distinction from adjacent domains. We further highlight key algorithms and approaches—including pruning, quantization, knowledge distillation, transfer learning, curriculum learning, and federated learning—that contribute to energy-aware model development. In addition, we examine tools and frameworks designed to measure carbon emissions and energy consumption, offering practical guidance for evaluating and minimizing the environmental footprint of AI systems. By synthesizing current progress and providing a structured taxonomy, this work aims to support the broader adoption of sustainable practices in AI research and development.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0