Table 1 Classification results on reduced STL-10 dataset without overlapping pooling
From: The effect of whitening transformation on pooling operations in convolutional autoencoders
Pooling size | Pooling type | Unwhitened | Whitened ( ε = 1) | Whitened ( ε = 0.1) | Whitened ( ε = 0.01) |
|---|---|---|---|---|---|
19 × 19 | Avg | 75.44% | 74.53% | 81.31% | 80.00% |
Max | 77.75% | 74.97% | 79.84% | 77.50% | |
Ada | 76.66% | 75.94% | 80.50% | 80.78% | |
5 × 5 | Avg | 72.53% | 73.53% | 78.38% | 75.97% |
Max | 75.34% | 74.25% | 77.56% | 74.56% | |
Ada | 73.06% | 73.59% | 77.87% | 75.81% |