Hot Paper - Boosting the Generalization Capability in Cross-Domain Few-shot Learning via Noise-enhanced Supervised Autoencoder

Boosting the Generalization Capability in Cross-Domain Few-shot Learning via Noise-enhanced Supervised Autoencoder

利用噪声增强型有监督自动编码器提高跨域Few-Shot学习的泛化能力
騒音増強型を利用して、自動エンコーダを監督し、ドメインを跨ぐFew-Shot学習の一般化能力を向上させる
소음 증강 형 을 이용 하여 자동 인 코딩 기 를 감독 하여 크로스 필드 Few-Shot 학습 의 일반화 능력 을 향상 시킨다
Mejora de la capacidad de generalización del aprendizaje de la tienda de few de dominio cruzado mediante el uso de codificador automático supervisado mejorado por ruido
Amélioration de la capacité de généralisation de l'apprentissage Multi - domaines par l'utilisation d'un encodeur automatique supervisé amélioré par le bruit
Повышение возможностей обобщения в междоменном обучении за несколько шагов с помощью контролируемого автоэнкодера с повышенным шумом

Hanwen Liang ¹, Qiong Zhang 张琼 ², Peng Dai ¹, Juwei Lu ¹

¹ Huawei Noah’s Ark Lab, Canada
加拿大华为诺亚方舟实验室
² Department of Statistics, University of British Columbia, Vancouver, Canada

https://arxiv.org/abs/2108.05028

arXiv, 11 August 2021

Abstract

State of the art (SOTA) few-shot learning (FSL) methods suffer significant performance drop in the presence of domain differences between source and target datasets. The strong discrimination ability on the source dataset does not necessarily translate to high classification accuracy on the target dataset.

In this work, we address this cross-domain few-shot learning (CDFSL) problem by boosting the generalization capability of the model. Specifically, we teach the model to capture broader variations of the feature distributions with a novel noise-enhanced supervised autoencoder (NSAE).

NSAE trains the model by jointly reconstructing inputs and predicting the labels of inputs as well as their reconstructed pairs. Theoretical analysis based on intra-class correlation (ICC) shows that the feature embeddings learned from NSAE have stronger discrimination and generalization abilities in the target domain. We also take advantage of NSAE structure and propose a two-step fine-tuning procedure that achieves better adaption and improves classification performance in the target domain.

Extensive experiments and ablation studies are conducted to demonstrate the effectiveness of the proposed method. Experimental results show that our proposed method consistently outperforms SOTA methods under various conditions.

Reviews and Discussions