EvoPrunerPool, an Evolutionary Pruner for Convolutional Neural Network compression, frames the process of filter pruning as a search problem to discover the optimal set of pruners from a selection of pre-existing filter pruners. These chosen pruners are then systematically applied in a specific sequence to gradually enhance the sparsity of a given Convolutional Neural Network.

EvoPrunerPool’s effectiveness has been showcased through experimentation on the LeNet model with MNIST data and the VGG-19 deep model with CIFAR-10 data. Its performance has been assessed against cutting-edge model compression techniques, establishing its benchmark status.

The paper was a collaborative effort by the MulticoreWare’s MAGIC cluster team at Amritha University. The paper was released by the Association of Computing Machinery (ACM).