Skip to main content

Exact Match Convolution

So, I tried to test out exact match convolution, which can be defined as:

It consists from three parts:
  • pixel offset
  • standard convolution
  • convolution bias
If you test it out with DeeperThought, with config:

convolutionsq,1,28,28,500,8,8,0.0,1.0,-0.001
max,500,21,21,21,21
matrix,500,200,0.1,-0.001
sigmoid,200
dropout,200,0.5
matrix,200,10,0.1,-0.001
softmax,10

where convolutionsq is exact match convolution (emc), you get pretty good results:



Top accuracy on test data: 98.68 % (Dataset - MNIST digits).

Not as good as with convolution (up to 99.41 % accuracy), but pretty close. Are we really just looking for exact matches and we do so by convolution? Hint: Linear Regression on a Set of Selected Templates from a Pool of Randomly Generated Templates.