TVM性能评估分析（四）

TVM性能评估分析（四）

 

 

 Figure 1.  Efficient Privacy-Preserving ML Using TVM

 

 

 Figure 2.  Motivation: Privacy-Preserving ML

 

 

 Figure 3.  Backend

 

 

  Figure 4. Differential privacy (DP) provides a formal guarantee that models trained on similar datasets are indistinguishable Informally

 

 

 Figure 5. System Overview

 

 

 Figure 6. Comparison of Traditional Auto-tuning and AutoTVM

 

 

   Figure 7.  ARM CPU

 

 

 Figure 8.  Mali GPU

 

 

 Figure 9. NVIDIA GPU

 

 

 Figure 10. AMD GPU

 

 

 Figure 11.  Automatic Kernel Optimization for Deep Learning on All Hardware Platforms