Feature-Space Generative Models for One-Shot Class-Incremental Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Foster, Jack, Paramonov, Kirill, Ozay, Mete, Michieli, Umberto
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912848512286720
author Foster, Jack
Paramonov, Kirill
Ozay, Mete
Michieli, Umberto
author_facet Foster, Jack
Paramonov, Kirill
Ozay, Mete
Michieli, Umberto
contents Few-shot class-incremental learning (FSCIL) is a paradigm where a model, initially trained on a dataset of base classes, must adapt to an expanding problem space by recognizing novel classes with limited data. We focus on the challenging FSCIL setup where a model receives only a single sample (1-shot) for each novel class and no further training or model alterations are allowed after the base training phase. This makes generalization to novel classes particularly difficult. We propose a novel approach predicated on the hypothesis that base and novel class embeddings have structural similarity. We map the original embedding space into a residual space by subtracting the class prototype (i.e., the average class embedding) of input samples. Then, we leverage generative modeling with VAE or diffusion models to learn the multi-modal distribution of residuals over the base classes, and we use this as a valuable structural prior to improve recognition of novel classes. Our approach, Gen1S, consistently improves novel class recognition over the state of the art across multiple benchmarks and backbone architectures.
format Preprint
id arxiv_https___arxiv_org_abs_2601_17905
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Feature-Space Generative Models for One-Shot Class-Incremental Learning
Foster, Jack
Paramonov, Kirill
Ozay, Mete
Michieli, Umberto
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Few-shot class-incremental learning (FSCIL) is a paradigm where a model, initially trained on a dataset of base classes, must adapt to an expanding problem space by recognizing novel classes with limited data. We focus on the challenging FSCIL setup where a model receives only a single sample (1-shot) for each novel class and no further training or model alterations are allowed after the base training phase. This makes generalization to novel classes particularly difficult. We propose a novel approach predicated on the hypothesis that base and novel class embeddings have structural similarity. We map the original embedding space into a residual space by subtracting the class prototype (i.e., the average class embedding) of input samples. Then, we leverage generative modeling with VAE or diffusion models to learn the multi-modal distribution of residuals over the base classes, and we use this as a valuable structural prior to improve recognition of novel classes. Our approach, Gen1S, consistently improves novel class recognition over the state of the art across multiple benchmarks and backbone architectures.
title Feature-Space Generative Models for One-Shot Class-Incremental Learning
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2601.17905