US 11,915,685 B2
Semi-sorted batching with variable length input for efficient training
Zhenhao Ge, Plano, TX (US); Lakshmish Kaushik, San Jose, CA (US); Saket Kumar, Fremont, CA (US); and Masanori Omote, San Mateo, CA (US)
Assigned to Sony Interactive Entertainment Inc., Tokyo (JP)
Filed by Sony Interactive Entertainment Inc., Tokyo (JP)
Filed on Mar. 23, 2023, as Appl. No. 18/189,145.
Application 18/189,145 is a continuation of application No. 17/107,362, filed on Nov. 30, 2020, granted, now 11,615,782.
Prior Publication US 2023/0326452 A1, Oct. 12, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G10L 13/02 (2013.01); G10L 15/06 (2013.01)
CPC G10L 15/063 (2013.01) [G10L 13/02 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A device comprising:
at least one computer storage that is not a transitory signal and that comprises instructions executable by at least one processor to:
randomly perturb respective numeric representations of lengths of respective machine learning (ML) training samples to establish respective pseudo-lengths;
sort the ML training samples by pseudo-length; and
train at least one ML model by inputting the ML training samples to the ML model.