US 11,995,540 B2
Online learning for dynamic Boltzmann machines with hidden units
Hiroshi Kajino, Tokyo (JP); and Takayuki Osogami, Kanagawa-ken (JP)
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on Oct. 11, 2018, as Appl. No. 16/157,455.
Prior Publication US 2020/0117987 A1, Apr. 16, 2020
Int. Cl. G06N 3/08 (2023.01); G06N 3/044 (2023.01); G06N 3/047 (2023.01)
CPC G06N 3/08 (2013.01) [G06N 3/044 (2023.01); G06N 3/047 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for online learning for a Dynamic Boltzmann Machine (DyBM) with hidden units, comprising:
imposing, by a processor device, limited connections in the DyBM where (i) a current observation depends only on an immediately preceeding two latest hidden units of a same layer connected to each other and all previous sequentially connected observations and (ii) the immediately preceding two latest hidden units depend on all the previous sequentially connected observations while lacking any connection with prior sequential hidden units situated directly before the immediately preceding two latest hidden units;
computing, by the processor device, gradients of an objective function of the DyBM; and
optimizing, by the processor device, the objective function in polynomial time with respect to the limited connections using a stochastic Gradient Descent algorithm applied to the gradients to provide a trained DyBM.