larochelle-metalearning.pdf英文版。很好的资源，适合机器学习以及人工智能爱

文件名称: larochelle-metalearning.pdf

所属分类: 机器学习

开发工具:

文件大小: 22mb

下载次数: 0

上传时间: 2019-07-29

提供者: xc***

下载 (22mb)

不能下载？报告错误

详细说明：英文版。很好的资源，适合机器学习以及人工智能爱好者。A RESEARCH AGENDA Deep learning successes have required a lot of labeled training data s collecting and labeling such data requires significant human labor practically, is that really how we'll solve Al sCientifically, this means there is a gap with ability of humans to learn, which we should try to understand Alternative solution: exploit other sources of data that are imperfect but plentiful s unlabeled data(unsupervised learning) multimodal data(multimodal learning multidomain data(transfer learning, domain adaptation) 2 3 5 ? D train Dtest People are good at it ⑤咱 Human-level concept learning People are through probabilistic good at it program induction Brenden m. lake.* Ruslan salakhutdinoy, Joshua B. Tenenbaum Machines are getting better at it 可aed Teachable machine ◎X 0aSecurehttps:/teachablemachine.withgoogle.com ☆员0●点 LEARNING O EXAMPLES CONFIDENCE OUTPUT INPUT GIF Sound Speech TRAIN GREEN O EXAMPLES CONFIDENCE TRAIN PURPLE O EXAMPLES CONFIDENCE TRAIN ORANGE Teachable machine ◎X 0aSecurehttps:/teachablemachine.withgoogle.com ☆员0●点 LEARNING O EXAMPLES CONFIDENCE OUTPUT INPUT GIF Sound Speech TRAIN GREEN O EXAMPLES CONFIDENCE TRAIN PURPLE O EXAMPLES CONFIDENCE TRAIN ORANGE A RESEARCH AGENDA Let's attack directly the problem of few-shot learning we want to design a learning algorithm a that outputs a good parameters 0 of a model M,when fed a small dataset Derain(xi, 1i)J2=1 Idea: lets learn that algorithm A, end-to-end this is known as meta-learning or learning to learn RELATED WORK TRANSFER LEARNING Large image datasets(e.g, ImageNet)have been shown to allow training representations that transfer to other problems DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition(20 14) Jeff Donahue, Yangqing ia, Oriol vinyals, Judy Hoffman, Ning zhang, Eric zeng and Trevor darrell CNN Features off-the-shelf: an Astounding Baseline for Recognition(2014 Ali sharif razavian, Hossein azizpour, osephine Sullivan, Stefan Carlsson some have even reported some positive transfer on medical imaging datasets In few-shot learning, We aim at transferring the complete training of the model on new datasets(not just transferring the features or initialization) s ideally there should be no human involved in producing a model for new datasets

(系统自动生成,下载前可以参看下载内容)