A Strong Baseline for Semi-Supervised Incremental Few-Shot Learning

TL;DR

This paper introduces a new semi-supervised incremental few-shot learning framework that effectively learns from limited labeled data and unlabeled data while maintaining performance on base classes.

Contribution

It proposes a novel paradigm with a meta-training algorithm and model adaptation mechanism for the complex S2 I-FSL setting, unifying semi-supervised and incremental FSL.

Findings

Effective in standard FSL, semi-supervised FSL, incremental FSL, and S2 I-FSL benchmarks.

Mitigates ambiguity caused by pseudo labels.

Preserves base class knowledge while learning new classes.

Abstract

Few-shot learning (FSL) aims to learn models that generalize to novel classes with limited training samples. Recent works advance FSL towards a scenario where unlabeled examples are also available and propose semi-supervised FSL methods. Another line of methods also cares about the performance of base classes in addition to the novel ones and thus establishes the incremental FSL scenario. In this paper, we generalize the above two under a more realistic yet complex setting, named by Semi-Supervised Incremental Few-Shot Learning (S2 I-FSL). To tackle the task, we propose a novel paradigm containing two parts: (1) a well-designed meta-training algorithm for mitigating ambiguity between base and novel classes caused by unreliable pseudo labels and (2) a model adaptation mechanism to learn discriminative features for novel classes while preserving base knowledge using few labeled and all the unlabeled data. Extensive experiments on standard FSL, semi-supervised FSL, incremental FSL, and the firstly built S2 I-FSL benchmarks demonstrate the effectiveness of our proposed method.