公开(公告)号：WO2021074743A1

公开(公告)日：2021-04-22

申请号：PCT/IB2020/059379

申请日：2020-10-06

Applicant: SISVEL TECHNOLOGY S.R.L. ,  INSTITUT MINES TÉLÉCOM

Inventor： TARTAGLIONE, Enzo ,  GRANGETTO, Marco ,  ODIERNA, Francesco ,  BRAGAGNOLO, Andrea ,  FIANDROTTI, Attilio

IPC: G06N3/04 ,  G06N3/08

Abstract: The present invention relates to a method for pruning a neural network (100) comprising a plurality of neurons (105), said method comprising:- an initialization phase, wherein input information is fetched comprising at least parameters ({wni,bni}) related to said neural network (100) and a dataset (D) representative of a task that said neural network (100) has to deal with, wherein said parameters ({wni,bni}) comprising a weights vector (wni) and/or a bias (bni) related to at least one neuron (105) of said plurality of neurons;- a regularization phase, wherein said neural network (100) is trained according to a training algorithm by using said dataset (D);- a thresholding phase, wherein an element (wnij) of said weights vector (wni) is put at zero when its absolute value is below a given threshold (T),said method wherein, during said regularization phase, said parameters ({wni,bni}) evolve according to a regularized update rule based on a neural sensitivity measure (S) to drive towards zero parameters related to at least one less sensitive neuron (108) of said neural network (100), wherein said neural sensitivity measure (S) is based on a pre-activation signal of at least one neuron (105) of said plurality of neurons.

公开(公告)号：EP4035087A1

公开(公告)日：2022-08-03

申请号：EP20789691.1

申请日：2020-10-06

Applicant: Sisvel Technology S.r.l. ,  Institut Mines Télécom

Inventor： TARTAGLIONE, Enzo ,  GRANGETTO, Marco ,  ODIERNA, Francesco ,  BRAGAGNOLO, Andrea ,  FIANDROTTI, Attilio

IPC: G06N3/04 ,  G06N3/08