Dl4j's AlexNet model interpretation based on the original paper ImageNet Classification with Deep Convolutional Neural Networks
and the imagenetExample code referenced.
Model is built in dl4j based on available functionality and notes indicate where there are gaps waiting for enhancements.
Bias initialization in the paper is 1 in certain layers but 0.1 in the imagenetExample code
Weight distribution uses 0.1 std for all layers in the paper but 0.005 in the dense layers in the imagenetExample code