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

---

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 (S2I-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. [Paper]

---

Highlights Contributions

A Semi-Supervised Incremental Few-Shot Learning (S2I-FSL) benchmark is proposed to generalize previous semi-supervised FSL and incremental FSL under a more realistic and challenging setting;

Technically, we propose an efficient meta-training paradigm and a model adaptation mechanism for the newly built task, which can be regarded as a baseline for future researches;

Extensive experiments on standard FSL, incremental FSL, semi-supervised FSL, and the S2I-FSL benchmark demonstrate the effectiveness of our proposed method

---

Recommended Citations

If you find our work is helpful to your research, please feel free to cite us:

		@inproceedings{zhao2021strong, 
			title={A Strong Baseline for Semi-Supervised Incremental Few-Shot Learning}, 
			author={Zhao, Linglan and Guo, Dashan and Xu, Yunlu and Qiao, Liang and Cheng, Zhanzhan and Pu, Shiliang and Niu, Yi and Fang, Xiangzhong}, 
			booktitle={BMVC},  
			year={2021}, 
		}