精华内容
下载资源
问答
  • Smart Multi-task BregmanClustering and Multi-task Kernel Clustering
  • multi-task

    2019-03-08 18:02:35
    Multi-task Learning Using Convolution-based Developmental Network PPT
  • Multi-modal Multi-task Learning for Automatic Dietary Assessment.
  • 1.multi-class, multi-label, multi-task classfication区别 每个标签有几个可能取值 几个标签 2 >2 1 binary classification multi-class classification >1 multi-label ...

    1. multi-class, multi-label, multi-task classfication区别

     每个标签有几个可能取值 
    几个标签2>2
    1binary classificationmulti-class classification
    >1multi-label classificationmulti-task classification

     

     

     

     

     

    2. 例子

    binary classification: 是不是食物是食物: (1)
    multi-class classification: 是食物?日用品?书本?是书本: (2) 或 (one hot: 0, 0, 1)
    multi-label classification: 是食物?塑料包装?甜的?是无包装的甜食: (1, 0, 1)
    multi-task classification: 是食物?塑料包装?出厂日期?是有包装的3月出厂的甜食: (1, 1, 2)


     

    展开全文
  • 一直很纠结Multi-class, Multi-label 以及 Multi-task 各自的区别和联系,最近找到了以下的说明资料: Multiclass classification means a classification task with more than two cla...

    一直很纠结Multi-class, Multi-label 以及 Multi-task 各自的区别和联系,最近找到了以下的说明资料:

    • Multiclass classification means a classification task with more than two classes; e.g., classify a set of images of fruits which may be oranges, apples, or pears. Multiclass classification makes the assumption that each sample is assigned to one and only one label: a fruit can be either an apple or a pear but not both at the same time.

    • Multilabel classification assigns to each sample a set of target labels. This can be thought as predicting properties of a data-point that are not mutually exclusive, such as topics that are relevant for a document. A text might be about any of religion, politics, finance or education at the same time or none of these.

    • Multioutput-multiclass classification and multi-task classification means that a single estimator has to handle several joint classification tasks. This is a generalization of the multi-label classification task, where the set of classification problem is restricted to binary classification, and of the multi-class classification task. The output format is a 2d numpy array or sparse matrix.

    The set of labels can be different for each output variable. For instance a sample could be assigned “pear” for an output variable that takes possible values in a finite set of species such as “pear”, “apple”, “orange” and “green” for a second output variable that takes possible values in a finite set of colors such as “green”, “red”, “orange”, “yellow”…

    This means that any classifiers handling multi-output multiclass or multi-task classification task supports the multi-label classification task as a special case. Multi-task classification is similar to the multi-output classification task with different model formulations. For more information, see the relevant estimator documentation.

    可以看出:

    • Multiclass classification 就是多分类问题,比如年龄预测中把人分为小孩,年轻人,青年人和老年人这四个类别。Multiclass classificationbinary classification相对应,性别预测只有男、女两个值,就属于后者。
    • Multilabel classification 是多标签分类,比如一个新闻稿A可以与{政治,体育,自然}有关,就可以打上这三个标签。而新闻稿B可能只与其中的{体育,自然}相关,就只能打上这两个标签。
    • Multioutput-multiclass classificationmulti-task classification 指的是同一个东西。仍然举前边的新闻稿的例子,定义一个三个元素的向量,该向量第1、2和3个元素分别对应是否(分别取值1或0)与政治、体育和自然相关。那么新闻稿A可以表示为[1,1,1],而新闻稿B可以表示为[0,1,1],这就可以看成是multi-task classification 问题了。 从这个例子也可以看出,Multilabel classification 是一种特殊的multi-task classification 问题。之所以说它特殊,是因为一般情况下,向量的元素可能会取多于两个值,比如同时要求预测年龄和性别,其中年龄有四个取值,而性别有两个取值。

    这里写图片描述

    展开全文
  • Discriminative multi-task multi-view feature selection and fusion for multimedia analysis
  • Joint Classification and Regression via Deep Multi-Task Multi-Channel Learning f
  • Multi-Task Sparse Canonical Correlation Analysis with Application to Multi-Modal Brain Imaging Genetics
  • KDD202论文M2GRL_A Multi-task Multi-view Graph Representation Learning Framework for Web-scale Recommender Systems | 图 | 推荐
  • multi-class,multi-label与multi-task的区别

    千次阅读 2018-06-10 13:52:18
    转自:https://blog.csdn.net/golden1314521/article/details/51251252一直很纠结Multi-class, Multi-label 以及 Multi-task 各自的区别和联系,最近找到了以下的说明资料:Multiclass classification means a ...

    转自:https://blog.csdn.net/golden1314521/article/details/51251252

    一直很纠结Multi-class, Multi-label 以及 Multi-task 各自的区别和联系,最近找到了以下的说明资料:

    • Multiclass classification means a classification task with more than two classes; e.g., classify a set of images of fruits which may be oranges, apples, or pears. Multiclass classification makes the assumption that each sample is assigned to one and only one label: a fruit can be either an apple or a pear but not both at the same time.

    • Multilabel classification assigns to each sample a set of target labels. This can be thought as predicting properties of a data-point that are not mutually exclusive, such as topics that are relevant for a document. A text might be about any of religion, politics, finance or education at the same time or none of these.

    • Multioutput-multiclass classification and multi-task classification means that a single estimator has to handle several joint classification tasks. This is a generalization of the multi-label classification task, where the set of classification problem is restricted to binary classification, and of the multi-class classification task. The output format is a 2d numpy array or sparse matrix.

      The set of labels can be different for each output variable. For instance a sample could be assigned “pear” for an output variable that takes possible values in a finite set of species such as “pear”, “apple”, “orange” and “green” for a second output variable that takes possible values in a finite set of colors such as “green”, “red”, “orange”, “yellow”…

      This means that any classifiers handling multi-output multiclass or multi-task classification task supports the multi-label classification task as a special case. Multi-task classification is similar to the multi-output classification task with different model formulations. For more information, see the relevant estimator documentation.

    可以看出:

    • Multiclass classification 就是多分类问题,比如年龄预测中把人分为小孩,年轻人,青年人和老年人这四个类别。Multiclass classification 与 binary classification相对应,性别预测只有男、女两个值,就属于后者。
    • Multilabel classification 是多标签分类,比如一个新闻稿A可以与{政治,体育,自然}有关,就可以打上这三个标签。而新闻稿B可能只与其中的{体育,自然}相关,就只能打上这两个标签。
    • Multioutput-multiclass classification 和 multi-task classification 指的是同一个东西。仍然举前边的新闻稿的例子,定义一个三个元素的向量,该向量第1、2和3个元素分别对应是否(分别取值1或0)与政治、体育和自然相关。那么新闻稿A可以表示为[1,1,1],而新闻稿B可以表示为[0,1,1],这就可以看成是multi-task classification 问题了。 从这个例子也可以看出,Multilabel classification是一种特殊的multi-task classification 问题。之所以说它特殊,是因为一般情况下,向量的元素可能会取多于两个值,比如同时要求预测年龄和性别,其中年龄有四个取值,而性别有两个取值。

    这里写图片描述

    展开全文
  • Multi-Task Multi-View Clustering读书笔记1

    千次阅读 2018-12-11 15:16:14
    这篇文章里只介绍CO-CLUSTERING BASED MULTI-TASK MULTI-VIEW CLUSTERING FRAMEWORK部分的内容 Multi-Task Multi-View Clustering是综合了Multi-TaskMulti-View两种方法来实现多任务学习,通过一个Common View...

    这篇文章里只介绍CO-CLUSTERING BASED MULTI-TASK MULTI-VIEW CLUSTERING FRAMEWORK部分的内容

    Multi-Task Multi-View Clustering是综合了Multi-Task和Multi-View两种方法来实现多任务学习,通过一个Common View筛选出样本的公共特征,利用公共特征来开展整个学习的过程,就我看来,文章里并没有提供Common View的计算方法,需要样本自己提供。

    Common View的概念如上图;

    CO-CLUSTERING BASED MULTI-TASK MULTI-VIEW CLUSTERING FRAMEWORK部分有以下几个部分。

    1.Within-View-Task Clustering

    2.Multi-View Relationship Learning

    3.Multi-Task Relationship Learning

    其中1是遍历每一个任务的每一个视角完成聚类,2的作用是使同一个任务下两两视角的差异之和最小,如何实现我将在后面提到。3是找出在同一个视角下相关任务的shared subspace。关于这个shared subspace,我的理解是公共样本矩阵和公共特征矩阵。

    接下来是算法的详细设计和构思

    1.Within-View-Task Clustering

    这个的证明是利用矩阵的迹的一些性质和U$^T$*U=I,U$^T$*M=I(MT是M的转置)以及D1,D2均为常量完成的,只需要将矩阵展开相乘再变化一下就可以由(2)——>(3),有时间我把证明过程贴上。

    1的主要目的是学习出合适的U、M,使(3)得以成立。

    矩阵转置的迹和原矩阵的迹相等这个就不用我说了吧。还有D1,D2是对角矩阵。

    2.Multi-View Relationship Learning

     

    这个的证明过程和上面的类似,都是利用tr()的性质变换矩阵,消去常量推导的·。他的目的是在一的基础上加入多视角方面的考虑,选择综合二者的优化后的 U$_t$$^{(v)}$M$_t$$^{(v)}$

    证明:

    3.Multi-Task Relationship Learning

    这一个利用的原理应该是SVD,通过common view v筛选X里的共享的那些特征(feature),然后利用SVD分解求出最优的共享特征的特征向量矩阵U$^{(v)}$,还有就是和1,2一起的M$_t$$^{(v)}$

    顺便说一下,

    U:the composed of the first c eigenvectors of features.U\in R$^{d*c}(d是特征数量,c是要聚类的样本数量)

    M:the composed of the first c eigenvectors of samples.U\in R$^{n*c}$(n是样本数量,c是要聚类的样本数量)

    二者分别是由X X$^T$计算出来的, U$_t$$^{(v)}$代表在task t&view v下U的子集。

    4.The Overall Objective Function

    从这里我们可以看出来,CO-CLUSTERING BASED MULTI-TASK MULTI-VIEW CLUSTERING FRAMEWORK是基于1的基础上纳入2,3中的multi-view的影响和multi-task的影响,得到需要的M$_t$$^{(v)}$,在它的基础上进行k-means算法,得到结果。

     

    计算优化:

    优化1

    其中,Definition1是找一个与目标矩阵最近似乎的矩阵X,满足X$^T$X=I (X不是方阵,所以不是正交矩阵,但有类似的意思,我也解释不清,希望大佬赐教)。Definition·2是把St(n,p)的东西变成函数\pi(X)。Proposition1给我们解释\pi(X)的计算方法。其余的就不多说了,(10)是梯度法学习参数+\pi(X)。顺带一提,这个方法的设计是利用参数之间的相互关系迭代优化设计的。给前一代的M$_t$$^{(v)}$,计算U$_t$$^{(v)}$

    优化2

    和优化1差不多。给M$_t$$^{(v)}$,算U$^{(v)}$(common view v下的特征矩阵)。

    优化3

    这个东西要分情况讨论,当M$_t$$^{(v)}$中的view v是common view 并且task t属于v下的公共任务集时,M$_t$$^{(v)}$需对整个目标式进行求解,因为我们目标函数的三个部分函数都有它的存在,见(14)。求解方法同优化1,优化2。反之,当view v不是common view 或者view v是common view ,但task t不属于v下的公共任务集时,Multi-Task Relationship Learning这一过程的条件它不满足,只计算前两部分。见(16)。

    Summary

    第一次写博客,做的不好的地方还有很多,如果有什么问题,请联系我,我会尽力解决。

    以上。

     

     

     

     

    展开全文
  • Multi-Task Learning as Multi-Objective Optimization 阅读笔记Multi-Task Learning(MTL)新的改变功能快捷键合理的创建标题,有助于目录的生成如何改变文本的样式插入链接与图片如何插入一段漂亮的代码片生成一个...
  • Nonparametric bayesian multi-task large-margin classification
  • including multi-task supervised learning, multi-task unsupervised learning, multi-task semi-supervised learning, multi-task active learning, multi-task reinforcement learning, multi-task online ...
  • Multi-task

    2014-05-13 20:51:27
    Feature Hashing for Large Scale Multitask Learning multitask learning with hundreds of thousands of tasks
  • Multi-Task Representation Learning for Demographic Prediction
  • Heterogeneous Multi-task Semantic Feature Learning forClassification
  • Multi-Task Model and Feature Joint Learning
  • Multi-Task Semi-Supervised Semantic Feature Learning for Classification
  • lec-16-Transfer and Multi-Task Learning.pdf
  • Object Localization via Evaluation on Multi-task Network
  • 论文题目: Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts 论文地址: https://dl.acm.org/citation.cfm?id=3220007 论文发表于: KDD 2018(CCF A类会议) 论文大体内容:...
  • Adaptive Multi-Task Transfer Learning for Chinese Word Segmentationin Medical Text英文文献,总结翻译在下一篇
  • 几篇CVPR关于multi-task的论文笔记整理,包括 一、 多任务课程学习Curriculum Learning of Multiple Tasks 1 --------------^CVPR2015/CVPR2016v--------------- 5 二、 词典对分类器驱动卷积神经网络进行对象检测...

空空如也

空空如也

1 2 3 4 5 ... 20
收藏数 42,821
精华内容 17,128
关键字:

multi-task