Most of the pruning algorithms depend on training data iterations to evaluate which synaptic weights can be pruned/removed to make the network optimal without affect performance on test data. New class of pruning algorithms have been developed which prune during initialization without looking at the data which however suffer from catastrophic layer-collapse or require an impractical amount of computation to obtain them. In this study, the authors claim to present a method that is (i) data independent, (ii) computationally efficiet, and (iii) achieves better performance to existing pruning algorithms.<p>Key idea of their iterative approach:<p>"..conservation alone leads to layer-collapse by assigning parameters in the largest layers with lower scores relative to parameters in smaller layers. However, if conservation is coupled with iterative pruning, then when the largest layer is pruned, becoming smaller, then in subsequent iterations the remaining parameters of this layer will be assigned higher relative scores. With sufficient iterations, conservation coupled with iteration leads to a self-balancing pruning strategy allowing IMP to avoid layer-collapse."