My code for Kaggle-MNIST competition.
Preprocessing | none |
Feature extraction | raw data |
Normalization | L2 normalization |
Classifier | Large Margin Component Analysis(LMCA)-based k-Nearest Neighbors |
Result | 0.98100 |
Preprocessing | 2px padding |
Feature extraction | Rectangle Features |
Normalization | Local standardization, ZCA Whitening, Global standardization |
Classifier | 2-Layer NN (2048 hidden units, Softmax, Dropout) |
Result | 0.98829 |