Vincenzo Laveglia (DIISM, Universities of Florence and Siena)
Oct 4, 2018 – 9:30 AM
DIISM, Artificial Intelligence laboratory (room 201), Siena SI
In this preliminary work we try to formalize the notion of sparsity in neural networks defining an appropriate indicator that we call sparsity index (SI), and investigate how this indicator evolves during the training process. Starting from this empirical analysis, we try to initialize the weights of a network (only the first layer actually) in order to have a desired SI.