Hardware and deep learning

1234567891011121314
Across
  1. 3. The place to store activations in a tiny device
  2. 6. Process of identify suitable CNN architecture
  3. 8. The non-dominated points during DSE
  4. 12. Using INT8 instead of FP32 representation
  5. 14. Determines the range of a floating point number
Down
  1. 1. An array structure used in most CNN accelerators
  2. 2. A common objective metric used in NAS
  3. 4. A CNN which proposed depthwise and pointwise convolutions
  4. 5. Helps map a CNN graph efficiently to given hardware
  5. 7. Removing unimportant weights from a CNN model
  6. 8. The layer which helps reduce the input dimensions for next layer
  7. 9. A commonly used bio-inspired algorithm for DSE
  8. 10. A lossless compression scheme used in Eyeriss
  9. 11. Accelerators help to reduce this metric
  10. 13. A device to implement custom digital logic