Convolutional neural network trained on the MNIST data set. The structure of the convolution is as follows:
1x28x28 Input -> 8x5x5 Convolution -> 2x2 Max Pooling -> 8x5x5 Convolution -> 2x2 Max Pooling -> 15 Hidden -> 10 Output
In the data folder, there are three weights files of various number of iterations.
File Name | "weights1.txt" | "weights2.txt" | "weights3.txt" |
---|---|---|---|
Number of Gradient Descent Iterations | 749529 | 604315 | 2644380 |
Test Case Accuracy | 96.31% | 96.36% | 96.39% |