Joselynzhao__Summer tree_CSDN博客-深度学习,论文解析,深度学习系统学习教程与实战案列领域博主

原创 openai docker 自定义基础容器服务器配置

基本模式是这样子的.我们需要将服务器上现成的image pull到本地, 按照自己的实验环境进行配置,保存为新的image, 然后再push到服务器上.假设服务器IP为72.13.0.0首先设置安全仓库为服务器Ipvim /etc/docker/daemon.json# 修改内容如下{"insecure-registries":["172.13.0.0"]}# 重启docker 使得配置...

2020-07-31 18:29:17 59

原创 docker入门课程

Docker 架构Ubuntu Docker 安装使用官方安装脚本自动安装手动安装Docker 镜像加速Docker Hello World运行交互式的容器启动容器（后台模式）停止容器Docker 容器使用获取镜像启动容器启动已停止运行的容器后台运行停止一个容器进入容器导出和导入容器删除容器Docker 镜像使用列出镜像列表获取一个新的镜像查找镜像删除镜像创建镜像更新镜像**构建镜像**设置镜像标签Docker 容器连接Docker 架构Docker 包括三个基本概念:镜像（Image）：Dock.

2020-07-30 14:25:04 36

原创【DKNN】Distilling the Knowledge in a Neural Network 第一次提出神经网络的知识蒸馏概念

Distilling the Knowledge in a Neural NetworkAbstract1 Introduction2 Distillation2.1 Matching logits is a special case of distillation3 Preliminary experiments on MNIST4 Experiments on speech recognition4.1 Results5 Training ensembles of specialists on ver.

2020-07-23 11:02:22 52

原创【DML】Deep Mutual Learning 深度相互学习，神经网络也可以协作

Deep Mutual LearningAbstract1. Introduction2. Related WorkModel DistillationCollaborative Learning3. Deep Mutual Learning3.1. Formulation3.2. Optimisation3.3. Extension to Larger Student Cohorts3.4. Extension to Semi-supervised Learning5. Conclusion论文：h.

2020-07-22 20:09:30 42

原创【MMT】ICLR 2020: MMT(Mutual Mean-Teaching)方法，无监督域适应在Person Re-ID上性能再创新高

为了减轻噪音伪标签的影响，文章提出了一种无监督的MMT（Mutual Mean-Teaching）方法，通过在迭代训练的方式中使用离线精炼硬伪标签和在线精炼软伪标签，来学习更佳的目标域中的特征。同时，还提出了可以让Traplet loss支持软标签的soft softmax-triplet loss”。该方法在域自适应任务方面明显优于所有现有的Person re-ID方法，改进幅度高达18.2％。MUTUAL MEAN-TEACHING: PSEUDO LABEL REFINERY FOR UNSUP.

2020-07-07 19:58:11 201

原创【FSR】Feature Space Regularization for Person Re-Identification with One Sample

Feature Space Regularization for Person Re-Identification with One SampleAbstractI. INTRODUCTIONFramework.Our Method.II. RELATED WORKSA. Supervised Re-IDB. Semi-supervised Re-IDC. Unsupervised re-IDD. Progressive LearningIII. THE PROPOSED METHODA. Overall.

2020-07-05 17:36:15 103 1

原创【CRR-FMM】A Concise Review of Recent Few-shot Meta-learning Methods

【CRR-FMM】A Concise Review of Recent Few-shot Meta-learning Methods1 IntroductionMindMap2. The Framework of Few-shot Meta-learning2.1. Notation and definitionsDefinition 1. (Small-sample learning)Definition 2. (Few-shot learning)Definition 3. (Few-shot met.

2020-05-29 19:16:30 137

原创少标签数据学习：宾夕法尼亚大学Learning with Few Labeled Data

文章目录Few-shot image classificationThree regimes of image classificationProblem formulationA flavor of current few-shot algorithmsHow well does few-shot learning work today?The key ideaTransductive LearningAn exampleResults on benchmark datasetsThe ImageNet.

2020-05-27 10:42:05 247

原创傅立叶的简单入门

参考文献：关于傅立叶的简单入门文章目录傅立叶的简单入门什么是傅立叶级数什么是傅立叶变换频域和时域傅立叶的简单入门将其中的信号表示成一组基本信号的线性组合，便于分析与观察输入和响应的关系来确定系统的特性。第一种分析方法是单位冲激响应的叠加形成的卷积。第二种就是通过傅立叶级数和傅立叶变换。更为深入…什么是傅立叶级数使LTI系统的信号表示成基本信号的组合，这些基本信号必须有两个性质：这些信号能构成相当广泛的一类信号LTI系统对这些信号的响应应该十分简单本身简单，衍生性还强第一幅图是

2020-05-15 09:47:52 148

原创卡耐基梅隆大学 Probabilistic Graphical Models 课程 | Elements of Meta-Learning 关于元学习和强化学习

Goals for the lecture:Introduction & overview of the key methods and developments.[Good starting point for you to start reading and understanding papers!]文章目录Probabilistic Graphical Models | Elements of Meta-Learning01 Intro to Meta-LearningMotiv.

2020-05-13 12:21:22 117

原创深度学习基础 Probabilistic Graphical Models | Statistical and Algorithmic Foundations of Deep Learning

文章目录Probabilistic Graphical ModelsStatistical and Algorithmic Foundations of Deep Learning01 An overview of DL componentsHistorical remarks: early days of neural networksReverse-mode automatic differentiation (aka backpropagation)Modern building blocks: u.

2020-05-12 18:46:57 125

原创最新小样本学习综述 A Survey on Few-Shot Learning | 四大模型Multitask Learning、Embedding Learning、External Memory…

文章目录01 Multitask Learning01.1 Parameter Sharing01.2 Parameter Tying.02 Embedding Learning相关阅读：A Survey on Few-Shot Learning | Introduction and OverviewA Survey of Few-Shot Learing | Data给定少数样本的Dt...

2020-05-11 10:56:59 613

原创 Generalizing from a Few Examples: A Survey on Few-Shot Learning 小样本学习最新综述 | 三大数据增强方法

文章目录01 Transforming Samples from Dtrain02 Transforming Samples from a Weakly Labeled or Unlabeled Data Set03 Transforming Samples from Similar Data SetsDiscussion and Summary上一篇：A Survey on Few-Shot ...

2020-04-29 15:53:18 434

原创 pycharm打开项目后只读变为可编辑

情况描述粘贴了一些文件到项目中, 打开pycharm准备编辑,发现文件属于只读模式,点击右下角的解锁按钮也无法编辑.解决方案命令格式:sudo chown -R username Filename输入密码后, 再次打开pycharm项目,便可以编辑了.参考链接: https://blog.csdn.net/weixin_34166472/article/details/863452...

2020-04-24 11:14:29 525

原创 linux系统能正常上网,但打不开github

问题linux系统能正常上网,但打不开github. 页面空白一直加载不出来内容.解决方法修改 /etc/hosts文件,记得给权限添加内容如下192.30.253.120 codeload.github.com192.30.253.113 github.com192.30.253.113 github.com192.30.253.118 gist.github.com192.3...

2020-04-17 09:56:53 764

原创 Generalizing from a Few Examples: A Survey on Few-Shot Learning 小样本学习最新综述| Introduction and Overview

Author listYAQING WANG, Hong Kong University of Science and Technology and Baidu ResearchQUANMING YAO∗, 4Paradigm Inc.JAMES T. KWOK, Hong Kong University of Science and TechnologyLIONEL M. NI, Ho...

2020-04-13 22:21:16 617

原创 Graph Neural Networks图神经网络(一)

Author: Nihai V. Nayak (March 2020)Graph Neural Networks图神经网络01 Introduction02 Basics03 Learning on Graphs03.1 Formal Definition04 Graph Convolutional Networks （GCN）04.1 Aggregate04.2 Combine05 Gr...

2020-04-12 22:32:36 559

原创 Capsule Networks胶囊网络（二）

文章目录Dynamic RoutingCoefficients operate on capsule levelDynamic Routing: Routing by AgreementComparison to fully connected neural networkfront-up contentComputing input/output vectors of a capsuleRout...

2020-04-12 21:13:06 141

原创仅需2步，利用GitHub Page生成网页

针对push到GitHub上面的web项目，利用Github Page生成网页下面举例说明博主已给GitHub page指定域名(joselynzhao.top), 没有域名的朋友，可以直接用{username}.github.io代替。项目目录如下：网站的index页面为/web/cws02.html01 在index file所在目录创建CMAKE文件内容为 Github...

2020-04-11 21:54:35 244

原创 Capsule Networks胶囊网络（一）

author: Sargur Srihari srihari@buffalo.eduThis is part of lecture slides on Deep Learning: http://www.cedar.buffalo.edu/~srihari/CSE676文章目录Limitations of Convolutional NetworksConvolutionalNeuralN...

2020-04-10 23:01:18 116

原创 Numpy （一）

Numpy （一）NumPy 官网 http://www.numpy.org/NumPy 源代码：https://github.com/numpy/numpySciPy 官网：https://www.scipy.org/SciPy 源代码：https://github.com/scipy/scipyMatplotlib 官网：https://matplotlib.org/Matplot...

2020-04-10 20:52:22 79

原创 Plant Leaves Classification: A Few-Shot Learning Method Based on Siamese Network

Plant Leaves Classification: A Few-Shot LearningMethod Based on Siamese NetworkAbstractIntroductionPROPOSED CNN STRUCTUREINITIAL CNN ANALYSISEXPERIMENTAL STRUCTURE AND ALGORITHMSAbstractIn this pap...

2019-11-24 19:35:29 14429

原创深度学习实验环境配置记录(ubuntu+docker+python+conda+pytorch)

文章目录ubuntu下安装docker安装验证docker 添加容器conda 容器配置conda 安装 pytorch实验项目准备附录ubuntu下安装docker参考链接 :菜根谭安装由于apt官方库里的docker版本可能比较旧，所以先卸载可能存在的旧版本sudo apt-get remove docker docker-engine docker-ce docker.io...

2019-11-22 19:31:13 804

原创 Learning Generalisable Omni-Scale Representations for Person Re-Identification

Learning Generalisable Omni-Scale Representations for Person Re-IdentificationAbstractIntroductionOSNet网络结构OMNI-SCALE NETWORK FOR PERSON RE-IDDepthwise Separable ConvolutionsOmni-Scale Residual Block实...

2019-10-31 21:44:35 515

原创基于深度学习的行人重识别研究综述

基于深度学习的行人重识别研究综述基于表征学习的ReID方法基于度量学习的ReID方法对比损失（Constrastive loss）三元组损失(Triplet loss)四元组损失(Quadruplet loss)难样本采样三元组损失(Triplet loss with batch hard mining, TriHard loss)Reference [原文]基于深度学习的行人重识别研究综述...

2019-10-29 15:42:51 970

原创 Decoupled Novel Object Captioner

Decoupled Novel Object CaptionerAbstractIntroductionMethodsPreliminariesZero-Shot Novel Object Captioning.Sequence Model with the PlaceholderKey-Value Object MemoryFramework OverviewReference（image-t...

2019-10-24 21:42:52 321

原创 Weakly Supervised Classification: Towards Accurate Machine Learning with Low Labeling Costs

Weakly Supervised Classification: Towards Accurate Machine Learning with Low Labeling Costs （Masashi Sugiyama:弱监督机器学习研究进展）报告摘要监督学习、非监督学习和半监督学习概述监督分类无监督分类半监督学习弱监督学习的研究进展PU数据分类PNU数据分类Pconf数据分类UU数据分类SU数据...

2019-10-24 15:33:58 310

原创 ZstGAN: An Adversarial Approach forUnsupervised Zero-Shot Image-to-Image Translation

ZstGAN: An Adversarial Approach forUnsupervised Zero-Shot Image-to-Image TranslationAbstractIntroductionMethodsProblem FormulationArchitectureReferenceAbstractIn this work In this workwe,we propose...

2019-10-24 10:44:54 369

原创 Generative Adversarial Text to Image Synthesis

Generative Adversarial Text to Image SynthesisAbstractIntroductionBackgroundGenerative adversarial networksMethodsReferenceAbstractIn this work, we develop a novel deeparchitecture and GAN formulati...

2019-10-23 17:08:09 219

原创 Dynamic Label Graph Matching for Unsupervised Video Re-Identification

Dynamic Label Graph Matching for Unsupervised Video Re-IdentificationAbstractIntroductionAbstractThis pa-per focuses on cross-camera label estimation, which can be subsequently used in feature learn...

2019-10-22 11:19:14 797

原创 Stepwise Metric Promotion for Unsupervised Video Person Re-identification

Stepwise Metric Promotion for Unsupervised Video Person Re-identificationAbstractIntroductionframeworkOur Method.ReferencesAbstracttwo assumptions two assumptionsdifferent video track-lets typical...

2019-10-21 21:28:17 219

原创 Image-Image Domain Adaptation with Preserved Self-Similarity andDomain-Dissimilarity for Person Re-i

Image-Image Domain Adaptation with Preserved Self-Similarity andDomain-Dissimilarity for Person Re-identificationmain methordsmain methordsIn our at-tempt, we present a “learning via translation” fr...

2019-10-21 20:46:38 399

原创 Re-ID 2019 Review

文章目录Train/Test on the same domainUnsupervised domain adaptationperson searchOther problemsFuture research questionsTrain/Test on the same domainPart-level features are effective.pose and 3D infor...

2019-10-19 17:34:37 308

原创为什么-关于因果关系的新科学 | 01 因果关系之梯

文章目录因果关系的三个层级迷你图灵测试论概率与因果关系上帝问的是“什么”,他们回答的却是“为什么”。上帝询问事实,他们回答理由。而且,两人都深信,列举原因可以以某种方式美化他们的行为。他们是从哪里得到这样的想法的?人类祖先想象不存在之物的能力是一切的关键,正是这种能力让他们得以交流得更加顺畅。在获得这种能力之前,他们只相信自己的直系亲属或者本部落的人。而此后,信任就因共同的幻想(例如信仰无...

2019-09-08 21:14:28 1319

原创为什么-关于因果关系的新科学 | 导言

2019-09-06 16:03:28 796

原创 ubuntu16.04下源码编译安装python3.6完成之后，ssl模块缺失无效，无法import ssl的问题的解决方法

ssl模块缺失表现为在python中执行如下命令import ssl会报错。在网上一查，是因为没有安装好openssl, 于是参照https://blog.csdn.net/xiaoqingshe0716/article/details/84886976　完成了openssl的源码安装。安装完成后，openssl的安装位置在/usr/local/openssl在终端输入openssl ...

2019-08-27 16:50:49 1637

原创 Zero-Shot Deep Domain Adaptation[reading notes]

文章目录AbstractIntroductionAbstractDomain adaptation(域适应) is an important tool to transfer knowledge about a task.Current approaches:假设 task-relevant target-domain数据在训练期间是可用的。而我们展示了如何在上述数据不可用的情况下实现...

2019-08-24 15:04:05 351

原创 Docker | 入门

什么是Docker?Docker是一个开源的引擎，可以轻松的为任何应用创建一个轻量级的、可移植的、自给自足的容器。开发者在笔记本上编译测试通过的容器可以批量地在生产环境中部署，包括VMs（虚拟机）、 bare metal、OpenStack 集群和其他的基础应用平台。Docker通常用于如下场景：web应用的自动化打包和发布；自动化测试和持续集成、发布；在服务型环境中部署和调整数据库或...

2019-08-20 20:40:46 128

原创深度学习 | 深度前馈网络

深度前馈网络前向网络隐藏层神经1 学习XOR2 基于梯度的学习代价函数输出单元用于高斯输出分布的线性单元基于伯努利输出分布的 sigmoid 单元用于multinouli 输出分布的softmax单元3 隐藏单元整流线性单元及其扩展4 架构设计5 反向传播和其他微分算法6 历史小计前向信息流过 x 的函数，流经用于定义 f 的中间计算过程，最终到达输出 y。网络前馈神经网络被称作网络 ...

2019-08-16 11:26:16 132

原创语法 | 01 定语从句

定语从句定语从句的基本知识先行词 and 关系词关系代词 and 关系副词限定 and 非限定从句的特征分析定语从句时的步骤关系代词引导的定语从句which 的特殊用法介词+关系代词定语从句的基本知识先行词 and 关系词定语从句的位置只能放在它修饰的词后面，被修饰的词叫做先行词。Tom，who is 17, is my classmate.Tom 先行词who 关系词先行词只...

2019-08-09 19:50:04 278