Vermeidung von Klassifizierungsfehler mit Hoher Kosten Beim Maschinenlernen

Details

Veröffentlichungs-Nr.

EP4027275

Aktenzeichen

EP21151138

Anmeldetag

12. Januar 2021

Veröffentlichung

13. Juli 2022

Rechtsraum

EP

IPC

G06N20/00G06N3/00G06F9/50G06Q10/00H04W4/02

Abstract

Disclosed is an apparatus comprising means for receiving, from a machine learning model, a classification of a first input data item, the machine learning model being trained to infer classifications of input data items, wherein determinations relating to performance of real-world actions are based at least in part on the classifications; responsive to the first input data item being classified by the machine learning model as a first classification: determining a robustness of the classification by the machine learning model of the first input data item; and responsive to a determination that the robustness of the classification of the first input data item as the first classification is below a first predefined robustness threshold robustness , outputting a second classification different to the first classification.

Anmelder

Firma

Nokia Solutions and Networks Oy

Land

🇫🇮 Finnland

Fachgebiete