Deep Learning · TEE · chapter

Convolutional neural networks

Convolution, pooling, output dimensions and parameter counts, including full network calculations.

  1. Reading image shapes before doing CNN arithmetic
  2. The convolution calculation
  3. Drill 34 Convolution with a negative coefficient
  4. Stride padding and floor
  5. Drill 35 Padding stride and channels
  6. Drill 38 Reverse a spatial formula
  7. One filter over multiple channels: all contributions go into one cell
  8. Pooling needs comparisons or averages
  9. Drill 36 Max and average pooling
  10. Parameters are different from activation values
  11. One complete CNN pipeline
  12. Drill 37 Complete small CNN
  13. Drill 46 Mixed convolution pool dense and loss
  14. Drill 49 Mixed CNN dimensions softmax and loss
  15. CNN versus a fully connected image network
  16. Faculty example: a complete binary-image convolution
  17. Faculty example: combine three RGB channel sums
  18. Max, min and average pooling on the same lecture grid
  19. Faculty CNN table: shapes, activations and 62958 parameters
  20. Even-sized kernels: SAME padding is not always symmetric
  21. F25 Q6a
  22. F25 Q7b
  23. R26 Q3b
  24. R26 Q6a
  25. F23 Q1d
  26. R24 Q1d
  27. R24 Q4b
  28. R24 Q5a
  29. F22 Q1a
  30. F22 Q3b
  31. F22 Q6a
  32. R23 Q1b
  33. R23 Q2a
  34. R23 Q4a
  35. R23 Q6b