Abstract:Recent work on few-shot learning \cite{tian2020rethinking} showed that quality of learned representations plays an important role in few-shot classification performance. On the other hand, the goal of self-supervised learning is to recover useful semantic information of the data without the use of class labels. In this work, we exploit the complementarity of both paradigms via a multi-task framework where we leverage recent self-supervised methods as auxiliary tasks. We found that combining multiple tasks is often beneficial, and that solving them simultaneously can be done efficiently. Our results suggest that self-supervised auxiliary tasks are effective data-dependent regularizers for representation learning. Our code is available at: \url{<a class="link-external link-https" href="https://github.com/nathanielsimard/improving-fs-ssl" rel="external noopener nofollow">this https URL</a>}.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to improve the performance of few - shot learning by introducing self - supervised learning auxiliary tasks**. Specifically, the author aims to prevent the model from overfitting to the base classes during the meta - training phase and improve the model's generalization ability to new classes by combining self - supervised learning methods as auxiliary tasks. ### Problem Background The goal of few - shot learning is to quickly learn new concepts from a limited number of samples, which is different from traditional image classification tasks that usually rely on a large amount of labeled data. Recent research has shown that the quality of the learned representation is crucial for few - shot classification performance (Tian et al., 2020a). Self - supervised learning, on the other hand, is dedicated to recovering useful semantic information of data without using class labels. ### Core Contributions of the Paper 1. **Combining Self - supervised Learning and Few - shot Learning**: The author proposes a multi - task framework, using recent self - supervised methods as auxiliary tasks to improve the performance of few - shot learning. 2. **Preventing Overfitting**: By introducing self - supervised auxiliary tasks, the model can learn more abundant feature representations during the meta - training process, thus avoiding overfitting to the base classes. 3. **Improving Generalization Ability**: The experimental results show that self - supervised auxiliary tasks can act as data - dependent regularizers, helping the model better generalize to new classes. ### Method Overview - **Problem Definition**: The paper adopts the standard N - way K - shot classification task to evaluate the model's performance. Each task contains a small number of classes and K training samples per class. - **Multi - task Learning Framework**: The author combines the supervised task with self - supervised tasks (such as rotation prediction and BYOL representation prediction), sharing the same feature extractor. The loss function is the sum of all tasks: \[ L_{\text{tot}}=\sum_{t = 1}^{N}L_t \] where \(L_t\) is the loss of the \(t\)-th task. - **Experimental Verification**: The author conducts experiments on two benchmark datasets, miniImageNet and CIFAR - FS. The results show that introducing self - supervised auxiliary tasks can significantly improve the performance of few - shot classification. ### Key Conclusions - Self - supervised auxiliary tasks (such as BYOL) can significantly improve the performance of few - shot learning. - Combining multiple self - supervised tasks can further improve the model's generalization ability. - Strengthening the data augmentation strategy also helps to improve the baseline performance of the supervised task. Through these improvements, the paper shows how to effectively combine self - supervised learning and few - shot learning to improve the model's generalization ability on new classes.

Improving Few-Shot Learning with Auxiliary Self-Supervised Pretext Tasks

Improving Few-shot Text Classification via Pretrained Language Representations

Three-Dimension Attention Mechanism and Self-Supervised Pretext Task for Augmenting Few-Shot Learning

Improving In-Context Few-Shot Learning via Self-Supervised Training

Self-Supervison with data-augmentation improves few-shot learning

Pareto Self-Supervised Training for Few-Shot Learning

Boosting Supervision with Self-Supervision for Few-shot Learning

When Low Resource NLP Meets Unsupervised Language Model: Meta-Pretraining then Meta-Learning for Few-Shot Text Classification (Student Abstract)

Less is More: A Closer Look at Semantic-based Few-Shot Learning

Few-Shot Image Classification via Contrastive Self-Supervised Learning

Semi-Supervised Few-Shot Classification With Multitask Learning and Iterative Label Correction

Task-Level Self-Supervision for Cross-Domain Few-Shot Learning

When Does Self-supervision Improve Few-Shot Learning?

TAFSSL: Task-Adaptive Feature Sub-Space Learning for few-shot classification

Self-Supervision Can Be a Good Few-Shot Learner

STraTA: Self-Training with Task Augmentation for Better Few-shot Learning

ESPT: A Self-Supervised Episodic Spatial Pretext Task for Improving Few-Shot Learning

Self-Supervised Task Augmentation for Few-Shot Intent Detection

SSL-ProtoNet: Self-supervised Learning Prototypical Networks for few-shot learning

Boosting Few-Shot Learning with Task-Adaptive Multi-level Mixed Supervision.