迁移学习（ Transfer Learning ）

转载于： http://apex.sjtu.edu.cn/apex_wiki/Transfer Learning

迁移学习（ Transfer Learning ）

薛贵荣

在传统的机器学习的框架下，学习的任务就是在给定充分训练数据的基础上来学习一个分类模型；然后利用这个学习到的模型来对测试文档进行分类与预测。然而，我们看到机器学习算法在当前的Web挖掘研究中存在着一个关键的问题：一些新出现的领域中的大量训练数据非常难得到。我们看到Web应用领域的发展非常快速。大量新的领域不断涌现，从传统的新闻，到网页，到图片,再到博客、播客等等。传统的机器学习需要对每个领域都标定大量训练数据，这将会耗费大量的人力与物力。而没有大量的标注数据，会使得很多与学习相关研究与应用无法开展。其次，传统的机器学习假设训练数据与测试数据服从相同的数据分布。然而，在许多情况下，这种同分布假设并不满足。通常可能发生的情况如训练数据过期。这往往需要我们去重新标注大量的训练数据以满足我们训练的需要，但标注新数据是非常昂贵的，需要大量的人力与物力。从另外一个角度上看，如果我们有了大量的、在不同分布下的训练数据，完全丢弃这些数据也是非常浪费的。如何合理的利用这些数据就是迁移学习主要解决的问题。迁移学习可以从现有的数据中迁移知识，用来帮助将来的学习。迁移学习（Transfer Learning）的目标是将从一个环境中学到的知识用来帮助新环境中的学习任务。因此，迁移学习不会像传统机器学习那样作同分布假设。

我们在迁移学习方面的工作目前可以分为以下三个部分：同构空间下基于实例的迁移学习，同构空间下基于特征的迁移学习与异构空间下的迁移学习。我们的研究指出，基于实例的迁移学习有更强的知识迁移能力，基于特征的迁移学习具有更广泛的知识迁移能力，而异构空间的迁移具有广泛的学习与扩展能力。这几种方法各有千秋。

1.同构空间下基于实例的迁移学习

基于实例的迁移学习的基本思想是，尽管辅助训练数据和源训练数据或多或少会有些不同，但是辅助训练数据中应该还是会存在一部分比较适合用来训练一个有效的分类模型，并且适应测试数据。于是，我们的目标就是从辅助训练数据中找出那些适合测试数据的实例，并将这些实例迁移到源训练数据的学习中去。在基于实例的迁移学习方面，我们推广了传统的AdaBoost算法，提出一种具有迁移能力的boosting算法：Tradaboosting [9]，使之具有迁移学习的能力，从而能够最大限度的利用辅助训练数据来帮助目标的分类。我们的关键想法是，利用boosting的技术来过滤掉辅助数据中那些与源训练数据最不像的数据。其中，boosting的作用是建立一种自动调整权重的机制，于是重要的辅助训练数据的权重将会增加，不重要的辅助训练数据的权重将会减小。调整权重之后，这些带权重的辅助训练数据将会作为额外的训练数据，与源训练数据一起从来提高分类模型的可靠度。

基于实例的迁移学习只能发生在源数据与辅助数据非常相近的情况下。但是，当源数据和辅助数据差别比较大的时候，基于实例的迁移学习算法往往很难找到可以迁移的知识。但是我们发现，即便有时源数据与目标数据在实例层面上并没有共享一些公共的知识，它们可能会在特征层面上有一些交集。因此我们研究了基于特征的迁移学习，它讨论的是如何利用特征层面上公共的知识进行学习的问题。

2.同构空间下基于特征的迁移学习

在基于特征的迁移学习研究方面，我们提出了多种学习的算法，如CoCC算法[7]，TPLSA算法[4]，谱分析算法[2]与自学习算法[3]等。其中利用互聚类算法产生一个公共的特征表示，从而帮助学习算法。我们的基本思想是使用互聚类算法同时对源数据与辅助数据进行聚类，得到一个共同的特征表示，这个新的特征表示优于只基于源数据的特征表示。通过把源数据表示在这个新的空间里，以实现迁移学习。应用这个思想，我们提出了基于特征的有监督迁移学习与基于特征的无监督迁移学习。

2.1 基于特征的有监督迁移学习

我们在基于特征的有监督迁移学习方面的工作是基于互聚类的跨领域分类[7]，这个工作考虑的问题是：当给定一个新的、不同的领域，标注数据及其稀少时，如何利用原有领域中含有的大量标注数据进行迁移学习的问题。在基于互聚类的跨领域分类这个工作中，我们为跨领域分类问题定义了一个统一的信息论形式化公式，其中基于互聚类的分类问题的转化成对目标函数的最优化问题。在我们提出的模型中，目标函数被定义为源数据实例，公共特征空间与辅助数据实例间互信息的损失。

2.2 基于特征的无监督迁移学习：自学习聚类

我们提出的自学习聚类算法[3]属于基于特征的无监督迁移学习方面的工作。这里我们考虑的问题是：现实中可能有标记的辅助数据都难以得到，在这种情况下如何利用大量无标记数据辅助数据进行迁移学习的问题。自学习聚类的基本思想是通过同时对源数据与辅助数据进行聚类得到一个共同的特征表示，而这个新的特征表示由于基于大量的辅助数据，所以会优于仅基于源数据而产生的特征表示，从而对聚类产生帮助。

上面提出的两种学习策略（基于特征的有监督迁移学习与无监督迁移学习）解决的都是源数据与辅助数据在同一特征空间内的基于特征的迁移学习问题。当源数据与辅助数据所在的特征空间中不同时，我们还研究了跨特征空间的基于特征的迁移学习，它也属于基于特征的迁移学习的一种。

３　异构空间下的迁移学习：翻译学习

我们提出的翻译学习[1][5]致力于解决源数据与测试数据分别属于两个不同的特征空间下的情况。在[1]中，我们使用大量容易得到的标注过文本数据去帮助仅有少量标注的图像分类的问题，如上图所示。我们的方法基于使用那些用有两个视角的数据来构建沟通两个特征空间的桥梁。虽然这些多视角数据可能不一定能够用来做分类用的训练数据，但是，它们可以用来构建翻译器。通过这个翻译器，我们把近邻算法和特征翻译结合在一起，将辅助数据翻译到源数据特征空间里去，用一个统一的语言模型进行学习与分类。

List of Conferences and Workshops Where Transfer Learning Paper Appear

From: http://www.cse.ust.hk/~sinnopan/conferenceTL.htm

List of Conferences and Workshops Where Transfer Learning Paper AppearThis webpage will be updated regularly.
Main Conferences
Machine Learning and Artificial Intelligence Conferences
AAAI	2008	Transfer Learning via Dimensionality Reduction [Link] [Bibtex]
		Transferring Localization Models across Space [Link] [Bibtex]
		Transferring Localization Models over Time [Link] [Bibtex]
		Transferring Multi-device Localization Models using Latent Multi-task Learning [Link] [Bibtex]
		Text Categorization with Knowledge Transfer from Heterogeneous Data Sources [Link] [Bibtex]
		Zero-data Learning of New Tasks [Link] [Bibtex]
	2007	Transferring Naive Bayes Classifiers for Text Classification [Link] [Bibtex]
		Mapping and Revising Markov Logic Networks for Transfer Learning [Link] [Bibtex]
		Measuring the Level of Transfer Learning by an AP Physics Problem-Solver [Link] [Bibtex]
	2006	Using Homomorphisms to Transfer Options across Continuous Reinforcement Learning Domains [Link] [Bibtex]
	2006	Value-Function-Based Transfer for Reinforcement Learning Using Structure Mapping [Link] [Bibtex]
IJCAI	2009	Transfer Learning Using Task-Level Features with Application to Information Retrieval [Link] [Bibtex]
		Transfer Learning from Minimal Target Data by Mapping across Relational Domains [Link] [Bibtex]
		Domain Adaptation via Transfer Component Analysis [Link] [Bibtex]
		Knowledge Transfer on Hybrid Graph [Link] [Bibtex]
		Manifold Alignment without Correspondence [Link] [Bibtex]
		Robust Distance Metric Learning with Auxiliary Knowledge [Link] [Bibtex]
		Can Movies and Books Collaborate? Cross-Domain Collaborative Filtering for Sparsity Reduction [Link] [Bibtex]
		Exponential Family Sparse Coding with Application to Self-taught Learning [Link] [Bibtex]
	2007	Learning and Transferring Action Schemas [Link] [Bibtex]
		General Game Learning Using Knowledge Transfer [Link] [Bibtex]
		Building Portable Options: Skill Transfer in Reinforcement Learning [Link] [Bibtex]
		Transfer Learning in Real-Time Strategy Games Using Hybrid CBR/RL [Link] [Bibtex]
		An Experts Algorithm for Transfer Learning [Link] [Bibtex]
		Transferring Learned Control-Knowledge between Planners [Link] [Bibtex]
		Effective Control Knowledge Transfer through Learning Skill and Representation Hierarchies [Link] [Bibtex]
		Efficient Bayesian Task-Level Transfer Learning [Link] [Bibtex]
ICML	2009	Deep Transfer via Second-Order Markov Logic [Link] [Bibtex]
		Feature Hashing for Large Scale Multitask Learning [Link] [Bibtex]
		A Convex Formulation for Learning Shared Structures from Multiple Tasks [Link] [Bibtex]
		EigenTransfer: A Unified Framework for Transfer Learning [Link] [Bibtex]
		Domain Adaptation from Multiple Sources via Auxiliary Classifiers [Link] [Bibtex]
		Transfer Learning for Collaborative Filtering via a Rating-Matrix Generative Model [Link] [Bibtex]
	2008	Bayesian Multiple Instance Learning: Automatic Feature Selection and Inductive Transfer [Link] [Bibtex]
		Multi-Task Learning for HIV Therapy Screening [Link] [Bibtex]
		Self-taught Clustering [Link] [Bibtex]
		Manifold Alignment using Procrustes Analysis [Link] [Bibtex]
		Automatic Discovery and Transfer of MAXQ Hierarchies [Link] [Bibtex]
		Transfer of Samples in Batch Reinforcement Learning [Link] [Bibtex]
		Hierarchical Kernel Stick-Breaking Process for Multi-Task Image Analysis [Link] [Bibtex]
		Multi-Task Compressive Sensing with Dirichlet Process Priors [Link] [Bibtex]
		A Unified Architecture for Natural Language Processing: Deep Neural Networks with Multitask Learning [Link] [Bibtex]
	2007	Boosting for Transfer Learning [Link] [Bibtex]
		Self-taught Learning: Transfer Learning from Unlabeled Data [Link] [Bibtex]
		Robust Multi-Task Learning with t-Processes [Link] [Bibtex]
		Multi-Task Learning for Sequential Data via iHMMs and the Nested Dirichlet Process [Link] [Bibtex]
		Cross-Domain Transfer for Reinforcement Learning [Link] [Bibtex]
		Learning a Meta-Level Prior for Feature Relevance from Multiple Related Tasks [Link] [Bibtex]
		Multi-Task Reinforcement Learning: A Hierarchical Bayesian Approach [Link] [Bibtex]
		The Matrix Stick-Breaking Process for Flexible Multi-Task Learning [Link] [Bibtex]
		Asymptotic Bayesian Generalization Error When Training and Test Distributions Are Different [Link] [Bibtex]
		Discriminative Learning for Differing Training and Test Distributions [Link] [Bibtex]
	2006	Autonomous Shaping: Knowledge Transfer in Reinforcement Learning [Link] [Bibtex]
	2006	Constructing Informative Priors using Transfer Learning [Link] [Bibtex]
NIPS	2008	Clustered Multi-Task Learning: A Convex Formulation [Link] [Bibtex]
		Multi-task Gaussian Process Learning of Robot Inverse Dynamics [Link] [Bibtex]
		Transfer Learning by Distribution Matching for Targeted Advertising [Link] [Bibtex]
		Translated Learning: Transfer Learning across Different Feature Spaces [Link] [Bibtex]
		An empirical Analysis of Domain Adaptation Algorithms for Genomic Sequence Analysis [Link] [Bibtex]
		Domain Adaptation with Multiple Sources [Link] [Bibtex]
	2007	Learning Bounds for Domain Adaptation [Link] [Bibtex]
		Transfer Learning using Kolmogorov Complexity: Basic Theory and Empirical Evaluations [Link] [Bibtex]
		A Spectral Regularization Framework for Multi-Task Structure Learning [Link] [Bibtex]
		Multi-task Gaussian Process Prediction [Link] [Bibtex]
		Semi-Supervised Multitask Learning [Link] [Bibtex]
		Gaussian Process Models for Link Analysis and Transfer Learning [Link] [Bibtex]
		Multi-Task Learning via Conic Programming [Link] [Bibtex]
		Direct Importance Estimation with Model Selection and Its Application to Covariate Shift Adaptation [Link] [Bibtex]
	2006	Correcting Sample Selection Bias by Unlabeled Data [Link] [Bibtex]
		Dirichlet-Enhanced Spam Filtering based on Biased Samples [Link] [Bibtex]
		Analysis of Representations for Domain Adaptation [Link] [Bibtex]
		Multi-Task Feature Learning [Link] [Bibtex]
AISTAT	2009	A Hierarchical Nonparametric Bayesian Approach to Statistical Language Model Domain Adaptation [Link] [Bibtex]
	2007	Kernel Multi-task Learning using Task-specific Features [Link] [Bibtex]
	2007	Inductive Transfer for Bayesian Network Structure Learning [Link] [Bibtex]
ECML/PKDD	2009	Relaxed Transfer of Different Classes via Spectral Partition [Link] [Bibtex]
		Feature Selection by Transfer Learning with Linear Regularized Models [Link] [Bibtex]
		Semi-Supervised Multi-Task Regression [Link] [Bibtex]
	2008	Actively Transfer Domain Knowledge [Link] [Bibtex]
		An Algorithm for Transfer Learning in a Heterogeneous Environment [Link] [Bibtex]
		Transferred Dimensionality Reduction [Link] [Bibtex]
		Modeling Transfer Relationships between Learning Tasks for Improved Inductive Transfer [Link] [Bibtex]
		Kernel-Based Inductive Transfer [Link] [Bibtex]
	2007	Graph-Based Domain Mapping for Transfer Learning in General Games [Link] [Bibtex]
		Bridged Refinement for Transfer Learning [Link] [Bibtex]
		Transfer Learning in Reinforcement Learning Problems Through Partial Policy Recycling [Link] [Bibtex]
		Domain Adaptation of Conditional Probability Models via Feature Subsetting [Link] [Bibtex]
	2006	Skill Acquisition via Transfer Learning and Advice Taking [Link] [Bibtex]
COLT	2009	Online Multi-task Learning with Hard Constraints [Link] [Bibtex]
		Taking Advantage of Sparsity in Multi-Task Learning [Link] [Bibtex]
		Domain Adaptation: Learning Bounds and Algorithms [Link] [Bibtex]
	2008	Learning coordinate gradients with multi-task kernels [Link] [Bibtex]
	2008	Linear Algorithms for Online Multitask Classification [Link] [Bibtex]
	2007	Multitask Learning with Expert Advice [Link] [Bibtex]
	2006	Online Multitask Learning [Link] [Bibtex]
UAI	2009	Bayesian Multitask Learning with Latent Hierarchies [Link] [Bibtex]
	2009	Multi-Task Feature Learning Via Efficient L2,1-Norm Minimization [Link] [Bibtex]
	2008	Convex Point Estimation using Undirected Bayesian Transfer Hierarchies [Link] [Bibtex]
Data Mining Conferences
KDD	2009	Cross Domain Distribution Adaptation via Kernel Mapping [Link] [Bibtex]
	2009	Extracting Discriminative Concepts for Domain Adaptation in Text Mining [Link] [Bibtex]
	2008	Spectral domain-transfer learning [Link] [Bibtex]
	2008	Knowledge transfer via multiple model local structure mapping [Link] [Bibtex]
	2007	Co-clustering based Classification for Out-of-domain Documents [Link] [Bibtex]
	2006	Reverse Testing: An Efficient Framework to Select Amongst Classifiers under Sample Selection Bias [Link] [Bibtex]
ICDM	2008	Unsupervised Cross-domain Learning by Interaction Information Co-clustering [Link] [Bibtex]
ICDM	2008	Using Wikipedia for Co-clustering Based Cross-domain Text Classification [Link] [Bibtex]
SDM	2008	Type-Independent Correction of Sample Selection Bias via Structural Discovery and Re-balancing [Link] [Bibtex]
	2008	Direct Density Ratio Estimation for Large-scale Covariate Shift Adaptation [Link] [Bibtex]
	2007	On Sample Selection Bias and Its Efficient Correction via Model Averaging and Unlabeled Examples [Link] [Bibtex]
	2007	Probabilistic Joint Feature Selection for Multi-task Learning [Link] [Bibtex]
Application Conferences
SIGIR	2009	Mining Employment Market via Text Block Detection and Adaptive Cross-Domain Information Extraction [Link] [Bibtex]
	2009	Knowledge transformation for cross-domain sentiment classification [Link] [Bibtex]
	2008	Topic-bridged PLSA for cross-domain text classification [Link] [Bibtex]
	2007	Cross-Lingual Query Suggestion Using Query Logs of Different Languages [Link] [Bibtex]
	2006	Tackling Concept Drift by Temporal Inductive Transfer [Link] [Bibtex]
		Constructing Informative Prior Distributions from Domain Knowledge in Text Classification [Link] [Bibtex]
		Building Bridges for Web Query Classification [Link] [Bibtex]
WWW	2009	Latent Space Domain Transfer between High Dimensional Overlapping Distributions [Link] [Bibtex]
WWW	2008	Can Chinese web pages be classified with English data source? [Link] [Bibtex]
ACL	2009	Transfer Learning, Feature Selection and Word Sense Disambiguation [Link] [Bibtex]
		Graph Ranking for Sentiment Transfer [Link] [Bibtex]
		Multi-Task Transfer Learning for Weakly-Supervised Relation Extraction [Link] [Bibtex]
		Cross-Domain Dependency Parsing Using a Deep Linguistic Grammar [Link] [Bibtex]
		Heterogeneous Transfer Learning for Image Clustering via the SocialWeb [Link] [Bibtex]
	2008	Exploiting Feature Hierarchy for Transfer Learning in Named Entity Recognition [Link] [Bibtex]
		Multi-domain Sentiment Classification [Link] [Bibtex]
		Active Sample Selection for Named Entity Transliteration [Link] [Bibtex]
		Mining Wiki Resources for Multilingual Named Entity Recognition [Link] [Bibtex]
		Multi-Task Active Learning for Linguistic Annotations [Link] [Bibtexs]
	2007	Domain Adaptation with Active Learning for Word Sense Disambiguation [Link] [Bibtex]
		Frustratingly Easy Domain Adaptation [Link] [Bibtex]
		Instance Weighting for Domain Adaptation in NLP [Link] [Bibtex]
		Biographies, Bollywood, Boom-boxes and Blenders: Domain Adaptation for Sentiment Classification [Link] [Bibtex]
		Self-Training for Enhancement and Domain Adaptation of Statistical Parsers Trained on Small Datasets [Link] [Bibtex]
	2006	Estimating Class Priors in Domain Adaptation for Word Sense Disambiguation [Link] [Bibtex]
	2006	Simultaneous English-Japanese Spoken Language Translation Based on Incremental Dependency Parsing and Transfer [Link] [Bibtex]
CVPR	2009	Domain Transfer SVM for Video Concept Detection [Link] [Bibtex]
	2009	Boosted Multi-Task Learning for Face Verification With Applications to Web Image and Video Search [Link] [Bibtex]
	2008	Transfer Learning for Image Classification with Sparse Prototype Representations [Link] [Bibtex]

Workshops

NIPS 2005 Workshop – Inductive Transfer: 10 Years Later

NIPS 2005 Workshop – Interclass Transfer

NIPS 2006 Workshop – Learning when test and training inputs have different distributions

AAAI 2008 Workshop – Transfer Learning for Complex Tasks

本文引用地址： http://www.sciencenet.cn/m/user_content.aspx?id=268960

迁移学习（ Transfer Learning ）相关推荐

迁移学习(Transfer learning)、重用预训练图层、预训练模型库
迁移学习(Transfer learning).重用预训练图层.预训练模型库目录迁移学习(Transfer learning).重用预训练图层.预训练模型库迁移学
深度学习不得不会的迁移学习Transfer Learning
http://blog.itpub.net/29829936/viewspace-2641919/ 2019-04-18 10:04:53 目录一.概述二.什么是迁移学习? 2.1 模型的训练与预 ...
迁移学习(transfer learning)与finetune的关系？【finetune只是transfer learning的一种手段】
目录 1.迁移学习简介 2.为什么要迁移学习? 3.迁移学习的几种方式 1)Transfer Learning: 2)Extract Feature Vector: 3)Fine-tune: 4.三种 ...
迁移学习(Transfer Learning)的背景、历史
迁移学习(Transfer Learning)的背景.历史迁移学习的背景.历史及学习 1.迁移学习提出背景在机器学习.深度学习和数据挖掘的大多数任务中,我们都会假设training和inferen ...
PyTorch 迁移学习 (Transfer Learning) 代码详解
PyTorch 迁移学习代码详解概述为什么使用迁移学习更好的结果节省时间加载模型 ResNet152 冻层实现模型初始化获取需更新参数训练模型获取数据完整代码概述迁移学习 ( ...
【深度学习系列】迁移学习Transfer Learning
在前面的文章中,我们通常是拿到一个任务,譬如图像分类.识别等,搜集好数据后就开始直接用模型进行训练,但是现实情况中,由于设备的局限性.时间的紧迫性等导致我们无法从头开始训练,迭代一两百万次来收敛模型, ...
迁移学习 Transfer Learning—通俗易懂地介绍（常见网络模型pytorch实现）
前言你会发现聪明人都喜欢"偷懒", 因为这样的偷懒能帮我们节省大量的时间, 提高效率. 还有一种偷懒是 "站在巨人的肩膀上". 不仅能看得更远, 还能看到更多 ...
迁移学习(Transfer Learning)概述及代码实现(full version)
基于PaddlePaddle的李宏毅机器学习--迁移学习大噶好,我是黄波波.希望能和大家共进步,错误之处恳请指出! 百度AI Studio个人主页, 我在AI Studio上获得白银等级,点亮2个徽 ...
迁移学习(Transfer Learning)概述及代码实现
基于PaddlePaddle的李宏毅机器学习--迁移学习大噶好,我是黄波波,希望能和大家共进步,错误之处恳请指出! 百度AI Studio个人主页, 我在AI Studio上获得白银等级,点亮2个徽 ...
【深度学习】迁移学习 (Transfer Learning)
Introduction 把别处学得的知识,迁移到新场景的能力,就是"迁移学习". 具体在实践中体现为: 将 A任务上预训练好的模型放在B任务上,加上少量B任务训练数据,进行微 ...

迁移学习（ Transfer Learning ）

迁移学习（ Transfer Learning ）相关推荐

最新文章

热门文章