EPO Patent Publication EP4711984A2: Self-supervised Representation Learning

ChangeBridge: EPO Bulletin - AI & Computing (G06N)

Published March 18th, 2026

Detected March 24th, 2026

Summary

The European Patent Office published patent application EP4711984A2 on March 18, 2026, detailing a method for self-supervised representation learning using bootstrapped latent representations. The patent, filed by GDM Holding LLC, describes a computer-implemented technique for training neural networks without labeled data or contrastive loss.

View original document View source feed page

What changed

The European Patent Office (EPO) has published patent application EP4711984A2, titled 'Self-supervised representation learning using bootstrapped latent representations.' This publication details a computer-implemented method for training neural networks, specifically an encoder neural network, without the need for labeled training data or contrastive loss (i.e., negative examples). The method involves processing transformed views of data with online and target neural networks, updating the online network's parameters to minimize error, and then updating the target network's parameters based on the online network's progress.

This publication is primarily of interest to technology companies and researchers in the field of artificial intelligence and machine learning. While it is a patent publication and not a regulatory rule, it signifies advancements in AI methodologies that could influence future technological development and intellectual property landscapes. Compliance officers should note the publication date and the specific AI techniques described, as patent filings can indicate emerging trends and potential future regulatory considerations in AI development and deployment.

Source document (simplified)

← EPO Patent Bulletin

SELF-SUPERVISED REPRESENTATION LEARNING USING BOOTSTRAPPED LATENT REPRESENTATIONS

Publication EP4711984A2 Kind: A2 Mar 18, 2026

Applicants

GDM Holding LLC

Inventors

GRILL, Jean-Bastien François Laurent, STRUB, Florian, ALTCHÉ, Florent, TALLEC, Corentin, RICHEMOND, Pierre, PIRES, Bernardo Avila, GUO, Zhaohan, AZAR, Mohammad Gheshlaghi, PIOT, Bilal, MUNOS, Remi, VALKO, Michal

Abstract

A computer-implemented method of training a neural network. The method comprises processing a first transformed view of a training data item, e.g. an image, with a target neural network to generate a target output, processing a second transformed view of the training data item, e.g. image, with an online neural network to generate a prediction of the target output, updating parameters of the online neural network to minimize an error between the prediction of the target output and the target output, and updating parameters of the target neural network based on the parameters of the online neural network. The method can effectively train an encoder neural network without using labelled training data items, and without using a contrastive loss, i.e. without needing "negative examples" which comprise transformed views of different data items.

IPC Classifications

G06N 3/084 20230101AFI20250818BHEP

Designated States

AL, AT, BE, BG, CH, CY, CZ, DE, DK, EE, ES, FI, FR, GB, GR, HR, HU, IE, IS, IT, LI, LT, LU, LV, MC, MK, MT, NL, NO, PL, PT, RO, RS, SE, SI, SK, SM, TR

View original document →