迁移学习(Transfer Learning)是一种机器学习方法,是把一个领域(即源领域)的知识,迁移到另外一个领域(即目标领域),使得目标领域能够取得更好的学习效果。迁移学习(TL)是机器学习(ML)中的一个研究问题,着重于存储在解决一个问题时获得的知识并将其应用于另一个但相关的问题。例如,在学习识别汽车时获得的知识可以在尝试识别卡车时应用。尽管这两个领域之间的正式联系是有限的,但这一领域的研究与心理学文献关于学习转移的悠久历史有关。从实践的角度来看,为学习新任务而重用或转移先前学习的任务中的信息可能会显着提高强化学习代理的样本效率。

知识荟萃

迁移学习荟萃20191209

综述

理论

模型算法

多源迁移学习

异构迁移学习

在线迁移学习

小样本学习

深度迁移学习

多任务学习

强化迁移学习

迁移度量学习

终身迁移学习

相关资源

领域专家

课程

代码

数据集

  • MNIST vs MNIST-M vs SVHN vs Synth vs USPS: digit images
  • GTSRB vs Syn Signs : traffic sign recognition datasets, transfer between real and synthetic signs.
  • NYU Depth Dataset V2: labeled paired images taken with two different cameras (normal and depth)
  • CelebA: faces of celebrities, offering the possibility to perform gender or hair color translation for instance
  • Office-Caltech dataset: images of office objects from 10 common categories shared by the Office-31 and Caltech-256 datasets. There are in total four domains: Amazon, Webcam, DSLR and Caltech.
  • Cityscapes dataset: street scene photos (source) and their annoted version (target)
  • UnityEyes vs MPIIGaze: simulated vs real gaze images (eyes)
  • CycleGAN datasets: horse2zebra, apple2orange, cezanne2photo, monet2photo, ukiyoe2photo, vangogh2photo, summer2winter
  • pix2pix dataset: edges2handbags, edges2shoes, facade, maps
  • RaFD: facial images with 8 different emotions (anger, disgust, fear, happiness, sadness, surprise, contempt, and neutral). You can transfer a face from one emotion to another.
  • VisDA 2017 classification dataset: 12 categories of object images in 2 domains: 3D-models and real images.
  • Office-Home dataset: images of objects in 4 domains: art, clipart, product and real-world.
  • DukeMTMC-reid and Market-1501: two pedestrian datasets collected at different places. The evaluation metric is based on open-set image retrieval.
  • Amazon review benchmark dataset: sentiment analysis for four kinds (domains) of reviews: books, DVDs, electronics, kitchen
  • ECML/PKDD Spam Filtering: emails from 3 different inboxes, that can represent the 3 domains.
  • 20 Newsgroup: collection of newsgroup documents across 6 top categories and 20 subcategories. Subcategories can play the role of the domains, as describe in this article.

实战


初步版本,水平有限,有错误或者不完善的地方,欢迎大家提建议和补充,会一直保持更新,本文为专知内容组原创内容,未经允许不得转载,如需转载请发送邮件至fangquanyi@gmail.com 或 联系微信专知小助手(Rancho_Fang)

敬请关注http://www.zhuanzhi.ai 和关注专知公众号,获取第一手AI相关知识

最近更新:2019-12-09

VIP内容

通过利用先前学习的任务来加速复杂任务的学习过程一直是强化学习中最具挑战性的问题之一,尤其是当源任务和目标任务之间的相似性较低时。本文针对深度强化学习中的知识迁移问题,提出了表示与实例迁移(REPAINT)算法。REPAINT 不仅在策略学习中转移了预先训练的教师策略的表示,而且还使用基于优势的经验选择方法来转移在非策略学习中按照教师策略收集的有用样本。本文在几个基准任务上的实验结果表明,在任务相似的一般情况下,REPAINT 显著减少了总训练时间。尤其是当源任务与目标任务不同或子任务不同时,REPAINT 在训练时间减少和返回分数的渐近表现方面都优于其他基线。

论文链接: https://www.zhuanzhi.ai/paper/0439c2852ae341fff43de69e5c7062ff

成为VIP会员查看完整内容
0
11

最新内容

Practical learning-based autonomous driving models must be capable of generalizing learned behaviors from simulated to real domains, and from training data to unseen domains with unusual image properties. In this paper, we investigate transfer learning methods that achieve robustness to domain shifts by taking advantage of the invariance of spatio-temporal features across domains. In this paper, we propose a transfer learning method to improve generalization across domains via transfer of spatio-temporal features and salient data augmentation. Our model uses a CNN-LSTM network with Inception modules for image feature extraction. Our method runs in two phases: Phase 1 involves training on source domain data, while Phase 2 performs training on target domain data that has been supplemented by feature maps generated using the Phase 1 model. Our model significantly improves performance in unseen test cases for both simulation-to-simulation transfer as well as simulation-to-real transfer by up to +37.3\% in test accuracy and up to +40.8\% in steering angle prediction, compared to other SOTA methods across multiple datasets.

0
0
下载
预览

最新论文

Practical learning-based autonomous driving models must be capable of generalizing learned behaviors from simulated to real domains, and from training data to unseen domains with unusual image properties. In this paper, we investigate transfer learning methods that achieve robustness to domain shifts by taking advantage of the invariance of spatio-temporal features across domains. In this paper, we propose a transfer learning method to improve generalization across domains via transfer of spatio-temporal features and salient data augmentation. Our model uses a CNN-LSTM network with Inception modules for image feature extraction. Our method runs in two phases: Phase 1 involves training on source domain data, while Phase 2 performs training on target domain data that has been supplemented by feature maps generated using the Phase 1 model. Our model significantly improves performance in unseen test cases for both simulation-to-simulation transfer as well as simulation-to-real transfer by up to +37.3\% in test accuracy and up to +40.8\% in steering angle prediction, compared to other SOTA methods across multiple datasets.

0
0
下载
预览
Top