US 11,989,653 B2
Pseudo-rounding in artificial neural networks
Gabriel Gouvine, Paris (FR)
Assigned to Mipsology SAS, Palaiseau (FR)
Filed by Mipsology SAS, Palaiseau (FR)
Filed on May 22, 2020, as Appl. No. 16/881,336.
Prior Publication US 2021/0365787 A1, Nov. 25, 2021
Int. Cl. G06N 3/08 (2023.01); G06F 17/16 (2006.01); G06N 3/04 (2023.01); G06N 3/082 (2023.01)
CPC G06N 3/082 (2013.01) [G06F 17/16 (2013.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A system comprising:
one or more processing units configured to:
receive a plurality of first parameters for one or more neurons of an artificial neural network (ANN), the first parameters being of a first data type;
generate, based on the plurality of the first parameters, an input in a format readable by a hardware accelerator, the hardware accelerator comprising a circuitry to perform computations of the one or more neurons of the ANN using a second data type, wherein:
the input includes a plurality of second parameters, each of the second parameters being a value of a set of values of the second data type;
the generation of the input includes changing the first parameters to the second parameters based on a distance between at least one first parameter and a corresponding second parameter being greater than a distance between the at least one first parameter and a value of the set of values of the second data type, the value being closest to the at least one first parameter; and
the plurality of the first parameters includes a parameter A, such that:
ÃL is a first number of the set of values of the second data type, ÃL being obtained by rounding the parameter A down;
ÃH is a second number of the set of values of the second data type, ÃH being obtained by rounding the parameter A up; and
the parameter A corresponds to a parameter à of the set of values of the second data type, the parameter à being located outside an interval [ÃL; ÃH]; and
provide the plurality of the second parameters to the hardware accelerator for computing the one or more neurons of the ANN.