mcordier/ML-neural-network-compression

e consider com-pression mechanism for neural network models proposed in [2]. In that paperthe kernel for L-ensembles process defining DPP used to subsample neuronsis fixed (Gaussian kernel). We experiment with different methods of choosingthis kernel and present our experimental results by comparing the quality ofthe obtained compressed neural networks. We want to beat DIVNETs fromthat paper. Authors : M.Cordier, JJ.Simeoni, T.Xu

Jupyter Notebook