Meta-learning with Diverse Tasks for Few-Shot Learning

ChangeBridge: Patent Apps - AI & Computing (G06N)

Published March 26th, 2026

Detected March 31st, 2026

Summary

USPTO published patent application US20260087412A1 by inventors Cresswell et al. covering methods for improving meta-learning models for few-shot learning of unseen tasks through improved task diversity scoring and generation of diverse training tasks using unsupervised analysis of disentangled latent features. Patent applications represent informational publications without creating immediate compliance obligations for third parties.

View original document View source feed page

What changed

This patent application discloses methods for training meta-learning models with improved task diversity for few-shot learning scenarios. The invention determines task diversity scores between pairs of tasks by partitioning a domain into respective classes, then scoring class pairs to determine similarity and subsequent task diversity. Diverse tasks are generated using unsupervised analysis of disentangled latent features, where each latent feature forms a task with classes based on clustering of data samples. The classes serve as training task labels for meta-learning model training.

This is a patent application publication only and does not create any compliance obligations or deadlines for regulated entities. Technology companies and AI researchers may monitor for subsequent patent grant notices that could affect intellectual property strategies, but no immediate action is required based on this publication alone.

Source document (simplified)

← USPTO Patent Applications

META-LEARNING WITH DIVERSE TASKS

Application US20260087412A1 Kind: A1 Mar 26, 2026

Inventors

Jesse Cole Cresswell, Yi Sui, Keyvan Golestan Irani, Maksims Volkovs, Wei Cui

Abstract

Meta-learning models are improved for few-shot learning of unseen tasks by improving task diversity of training data used for training the meta-learning model. A task diversity score may be determined between a pair of tasks that partition a domain into respective classes. The respective classes are paired and scored to determine similarity between class pairs and subsequent task diversity scores. Diverse tasks may be generated with unsupervised analysis of the domain by determining disentangled latent features of the data samples. Each latent feature may then be considered a task with classes based on a clustering of the data samples based on the feature values of the respective latent feature. The classes are then used as training task labels for the data samples and sampled from to generate diverse tasks for the meta-learning model.