MNIST
Problem Statement
Get at least 98% accuracy on the MNIST test set.
Rules
- Can only train the model on the MNIST training set
- Can use image transformations but input image size should remain the same (28x28)
- Model result must be replicatable
- Entire training process should be automated
- Optimiser weights aren't counted towards the total number of trainable parameters
Dataset
Leaderboard
| Rank | Name | Parameter Count |
|---|---|---|
| 1 | HexHowells | 575 |
| 2 | ||
| 3 | ||
| 4 | ||
| 5 | ||
| 6 | ||
| 7 | ||
| 8 | ||
| 9 | ||
| 10 |