注意下面很多链接需要访问外国网站,无奈国情如此
1. TensorFlow开发者峰会昨天召开,我觉得有趣的演讲有
- TF Hub ("一键"transfer learning)
- Swift支持 (Chris Lattner原来去Google Brain就是做这个)
- tensorflow.js (deeplearn.js升级版)
- performance tuning (这个部分我觉得讲的特别好)
- TensorBoard Debugger (wow,牛B的工具!)
- TF Lite (TF Mobile会被淘汰)
- AutoML (还是NAS那一套)
- TFX (生产系统,TF Transform很不错,Serving RESTful API)
视频:https://www.youtube.com/playlist?list=PLQY2H8rRoyvxjVx3zfw4vA4cvlKogyLNN
2. DeepMind用reinforced adversarial learning让机器学写程序来画画,跟GAN有点像
Learning to write programs that generate images
链接:https://deepmind.com/blog/learning-to-generate-images/
3. 分析Uber自驾车撞人事件的分析
Could AI have saved the cyclist (had I programmed the Uber car)?
链接:https://medium.com/@rebane/could-ai-have-saved-the-cyclist-had-i-programmed-the-uber-car-6e899067fefe
4. AutoML的资源收集,architectural search部分比较全
链接:http://www.ml4aad.org/automl/
4*. meta-learning解释
Learning About Algorithms That Learn to Learn
链接:https://towardsdatascience.com/learning-about-algorithms-that-learn-to-learn-9022f2fa3dd5
5. Berkeley/OpenAI用soft Q-learning做机器人学习
Reinforcement Learning with Deep Energy-Based Policies
链接:https://arxiv.org/pdf/1702.08165.pdf
6. PyTorch的RL实现教程
DQN Adventure: from Zero to State of the Art
链接:https://github.com/higgsfield/RL-Adventure?utm_campaign=Revue%20newsletter&utm_medium=Newsletter&utm_source=The%20Wild%20Week%20in%20AI
7. adversarial patch的论文分析(Morning Paper这个系列挺好的,分析很多CS的最新论文,虽然很多不是机器学习)
链接:https://blog.acolyer.org/2018/03/29/adversarial-patch/
8. deep learning这本书的解读,虽然暂时只有前面2章,但是不得不说visualization做的很漂亮
链接:https://hadrienj.github.io/posts/Deep-Learning-Book-Series-Introduction/
9. VAE最清楚的解释
Intuitively Understanding Variational Autoencoders
链接:https://towardsdatascience.com/intuitively-understanding-variational-autoencoders-1bfe67eb5daf
10. 工具/框架
10.1 阿里用TVM加速TF
Bringing TVM into TensorFlow for Optimizing Neural Machine Translation on GPU
链接:http://www.tvmlang.org/2018/03/23/nmt-transformer-optimize.html
10.2 Linux Foundation也作了一个深度学习的项目Acumos,看起来主要目的是整合各个framework
链接:https://www.acumos.org/
10.3 DL framework对比与选择,写的还算比较全面
链接:http://engineering.curalate.com/2018/03/23/DL-lib-for-app-dev-and-prod.html?utm_campaign=Revue%20newsletter&utm_medium=Newsletter&utm_source=Deep%20Learning%20Weekly
10.4 Javascript的GPU加速DL framework
链接:https://towardsdatascience.com/gpu-accelerated-neural-networks-in-javascript-195d6f8e69ef
10.5 神经网络可视化工具Netron
链接:https://github.com/lutzroeder/Netron
10.6 TF 1.7发布
链接:https://github.com/tensorflow/tensorflow/releases/tag/v1.7.0
本文分享自 机器学习人工学weekly 微信公众号,前往查看
如有侵权,请联系 cloudcommunity@tencent.com 删除。
本文参与 腾讯云自媒体同步曝光计划 ,欢迎热爱写作的你一起参与!