Kollaboratives Training Fairer Maschinenlernmodelle

Details

Veröffentlichungs-Nr.

EP4752785

Anmeldetag

19. Dezember 2024

Veröffentlichung

3. Juni 2026

Rechtsraum

EP

IPC

G06N3/045, G06N3/098

Abstract

A collaboratively trained model obtains an embedding vector and outputs an output vector. A first (second) classification head: (i) obtains an output vector element values; and (ii) outputs a predicted value of a first (second) attribute based on the output vector element values. The first (second) classification head has been trained by a first (second) entity to predict a value of the first (second) attribute based on the output vector element values. The predicted value of the first (second) attribute is useable by the first (second) entity to determine an accuracy of the first (second) classification head in predicting values of the first (second) attribute from output vector element values output by the model. The model is updated to increase (decrease) the accuracy of the first (second) classification head in predicting values of the first (second) attribute from output vector element values output by the model.

Anmelder

Firma

Featurespace Limited

Land

🇬🇧 Großbritannien