MNIST

Every data scientist eventually tackles MNIST. Here's my attempts.

1. (0.75171) LeNet set to minimize cross-entropy. Trained for 100 epochs.
2. (0.98614) AlexNet with the same parameters.
3. (0.98914) Modified GoogleNet with Batch Normalization instead of LRN, and an average pooling layer removed.
4. (0.98743) Googlenet from 3 with adamax optimizer used.
5. (Running) AlexNet with a grid search for parameters

Uploaded Models

Selected models have been uploaded in case anyone wants to play with them.