Open dorimedini opened 5 years ago
ATM pruning is done via a fixed constant threshold. We should prune edges layer by layer, each layer with a specific threshold original_threshold * l2_norm(layer_edge_weights)
original_threshold * l2_norm(layer_edge_weights)
ATM pruning is done via a fixed constant threshold. We should prune edges layer by layer, each layer with a specific threshold
original_threshold * l2_norm(layer_edge_weights)