Survey Maps Green Development for Large AI Models

The rapid growth of large-scale AI models has brought about significant performance gains but also raised serious concerns regarding their computational cost, energy consumption, and environmental impact. A new survey provides an extensive overview of efforts towards the 'green development' of these models. The survey systematically examines recent progress in creating more efficient models, including optimizations for attention mechanisms, linear-complexity architectures, and techniques like model sparsification and merging. It also delves into sustainable training and deployment strategies, such as data-efficient learning, parameter-efficient fine-tuning, and computational compression. Beyond algorithmic improvements, the review explores energy-efficient AI hardware, memory optimization, cross-platform deployment, and sustainable infrastructure. Furthermore, the paper highlights how large models are being applied to critical sustainability challenges, from remote sensing interpretation to national-scale infrastructure and global initiatives. It concludes by discussing key challenges and future research directions, emphasizing the need for continual learning paradigms, memory-centric hardware, and standardized evaluation protocols to ensure sustainable, scalable, and socially responsible AI development.