← Back to homepage

Green AI and Efficient Deep Learning

Research note · Sustainable AI · Work in progress

These notes collect concepts and practical ideas related to Green AI, efficient deep learning, and energy-aware model evaluation.

1. Why efficiency matters

Deep learning models are often evaluated mainly through predictive performance. However, training and inference also have computational, energy, and environmental costs. A more complete evaluation should consider both accuracy and resource usage.

2. Typical efficiency levers

• Reducing model size through pruning, distillation, quantization, or low-rank approximations.
• Choosing architectures adapted to the deployment environment.
• Reporting training and inference cost alongside standard performance metrics.
• Studying scaling laws and cost-performance trade-offs.

3. Open questions

Important questions include how to compare models fairly across hardware, how to estimate energy consumption reliably, and how to design models that are both accurate and resource-aware.