当前位置：网站首页>-2021 sorting and sharing of the latest required papers related to comparative learning

-2021 sorting and sharing of the latest required papers related to comparative learning

2022-07-23 21:34:00 【lqfarmer】

When it comes to comparative learning （Contrastive Learning）, First of all, we should start with self supervised learning . Self supervised learning is a kind of unsupervised learning paradigm , It is characterized by category label information that does not need to be marked manually , Directly use the data itself as supervision information , To learn the feature expression of sample data , And used for downstream tasks .

At present, self supervised learning can be roughly divided into two categories ：

Generative Learning vs Contrastive Learning

Generative Learning（ Generative approach ） This kind of method is represented by self encoder , Main concern pixel label Of loss. for instance , In the self encoder, the data samples are encoded into features and then decoded and reconstructed , Here, it is considered that if the reconstruction effect is better, it shows that the model has learned better feature expression , The effect of reconstruction is through pixel label Of loss To measure .

Contrastive Learning（ Comparative method ） This kind of method is to compare the data with positive samples and negative samples in the feature space , To learn the feature representation of the sample .Contrastive Learning The main difficulty is how to construct positive and negative samples .

As a new popular deep learning algorithm, contrastive learning , From the beginning, programming has gradually become the mainstream of academic and industrial research . This resource arranges the papers related to comparative learning in recent years , A friend in need takes it by himself .

Resources are organized from the Internet , See source address for resource acquisition ：

https://github.com/asheeshcric/awesome-contrastive-self-supervised-learning

List of papers