green-ai

Xu et al. 2021: A survey on green deep learning

A survey on various approaches for greener deep learning.

• Categories of approaches for green deep learning technologies:
  • Compact networks
  • Energy-efficient training strategies
    • Pre-trained models for initialization
      • Efficient hyper-parameter optimization
  • Energy-efficient inference approaches
    • Pruning
    • Low-rank factorization
    • Quantization
    • Knowledge distillation
  • Efficient data usage
• Future work:
  • Deploying models on edge
  • Combining edge and clode: Dynamic network where edge devices handle simple samples with smaller models, and the cloud is responsible for handling complicated samples with larger models.