Feedforward Networks on CIFAR-10

PrototypeUniversity

FFNN implementation with per-class evaluation and weight-distribution analysis. With Zuzanna Bojarska.

Training run: 3 layer MLP, CIFAR-10Sample data

Test accuracy52.4%

Train accuracy61.8%

Params1.6M

Epochs40

accuracy %epoch

The gap opening after epoch 20 is the point the write-up discusses.

PrototypeAn interface sketch of this system, not a screenshot. Every figure, name and record in it is invented.