Note：[ wechat：Y466551 | 可加勿骚扰，付费咨询 ]

论文信息

论文标题：FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence
论文作者：
论文来源：2020 aRxiv
论文地址：download 
论文代码：download
视屏讲解：click

1-介绍 

　　动机：解决现有半监督方法利用未标记数据来提高模型性能存在复杂度较高的问题；

2-相关

一致性正则化

　　一致性正则化利用未标记的数据，依赖于假设模型应该输出类似的预测时，输入相同的图像。

　　　　$\sum_{b=1}^{\mu B}\left\|p_{\mathrm{m}}\left(y \mid \alpha\left(u_{b}\right)\right)-p_{\mathrm{m}}\left(y \mid \alpha\left(u_{b}\right)\right)\right\|_{2}^{2}$

伪标签

　　伪标记利用了使用模型本身来获得未标记数据[32,47]的人工标签的想法。

　　　　$\frac{1}{\mu B} \sum_{b=1}^{\mu B} \mathbb{1}\left(\max \left(q_{b}\right) \geq \tau\right) \mathrm{H}\left(\hat{q}_{b}, q_{b}\right)$

　　思想：伪标记——熵最小化——低密度分离

3-方法

模型框架

算法

4-实验

　　略

本栏目推荐文章
• Early lameness detection in dairy cattle based on wearable gait analysis using semi-supervised LSTM-Autoencoder
• Graph regularized non-negative matrix factorization with prior knowledge consistency constraint for drug-target interactions prediction
• 简化版Transformer ：Simplifying Transformer Block论文详解
• 简化版Transformer ：Simplifying Transformer Block论文详解
• CA-TCC: 半监督时间序列分类的自监督对比表征学习《Self-supervised Contrastive Representation Learning for Semi-supervised Time-Series Classification》(时间序列、时序表征、时间和上下文对比、对比学习、自监督学习、半监督学习、TS-TCC的扩展版)
• Confidence First: Yale University's ideal of mind
• consistency level of Azure Cosmos DB account
• 基于时间频率一致性对时间序列进行自监督对比预训练《Self-Supervised Contrastive Pre-Training for Time Series via Time-Frequency Consistency》(时序、时频一致性、对比学习)
• Microservice - Data Consistency
• [论文阅读] Latent Consistency Models@ Synthesizing High-Resolution Images with Few-Step Inference

FixMatch Semi-Supervised Consistency Simplifying Confidencefixmatch semi-supervised consistency simplifying semi-supervised consistency adaptation supervised semi-supervised regularization consistency revisiting fixmatch confidence simplifying dbc-forest confidence screening reading recommendations distributions lightweight confidence confidence university first ideal transformer simplifying论文block