换了个新职位非常忙,我还会每周更新,但是估计是没时间仔细读论文了,所以评论可能就没什么深度了,精力有限,哎。
1. DeepMind新RL学习方式SAC-X,尝试通过auxiliary task来解决sparse reward问题(跟UNREAL有些关系)
Learning by playing
blog:https://deepmind.com/blog/learning-playing/
链接:https://arxiv.org/pdf/1802.10567.pdf
2. Google Brain通过增加auxiliary loss的方法改进LSTM
Learning Longer-term Dependencies in RNNs with Auxiliary Losses
链接:https://arxiv.org/abs/1803.00144
3. 阿里发布用多agent RL提高广告收益的论文
Deep Reinforcement Learning for Sponsored Search Real-time Bidding
链接:https://arxiv.org/abs/1803.00259
4. variational autoencoders讲的最清楚的视频(Arxiv Insights这个账号值得follow)
链接:https://www.youtube.com/watch?utm_campaign=Revue+newsletter&utm_medium=Newsletter&utm_source=The+Wild+Week+in+AI&v=9zKuYvjFFS8
5. Intel MobileEye发的一篇论文,把很多自驾车的安全问题用数学方法表示出来
On a Formal Model of Safe and Scalable Self-driving Cars
链接:https://arxiv.org/pdf/1708.06374.pdf
6. NLP的前沿研究问题,都是NLP的高级问题
链接:http://ruder.io/requests-for-research/?utm_campaign=Revue%20newsletter&utm_medium=Newsletter&utm_source=The%20Wild%20Week%20in%20AI
7. 框架/工具/资源
7.1 Berkeley RISE实验室的Panda on Ray,用并行化加速Panda处理数据
链接:https://rise.cs.berkeley.edu/blog/pandas-on-ray/
7.2 instacart开源lore,把整个机器学期的pipeline都包进去
链接:https://github.com/instacart/lore
简单教程:https://tech.instacart.com/how-to-build-a-deep-learning-model-in-15-minutes-a3684c6f71e
7.3 Alpha Zero Chess
链接:https://github.com/Zeta36/chess-alpha-zero
7.4 TF教程(有视频),看起来不错
链接:https://github.com/Hvass-Labs/TensorFlow-Tutorials
7.5 word vector教程
链接:https://gist.github.com/aparrish/2f562e3737544cf29aaf1af30362f469
7.6 微软发布MMdnn,让各个框架之间的模型互相转化,感觉跟ONNX有点重叠啊
链接:https://github.com/Microsoft/MMdnn
7.7 用keras实现各种GAN
链接:https://github.com/eriklindernoren/Keras-GAN?utm_campaign=Revue%20newsletter&utm_medium=Newsletter&utm_source=The%20Wild%20Week%20in%20AI
7.8 TF 1.6发布了
链接:https://github.com/tensorflow/tensorflow/releases
*Waymo发布VR视频,里面很清楚的可以看到LiDAR/radar/camera的perception,sensor fusion, planning, mapping等等
链接:https://www.youtube.com/watch?time_continue=70&v=B8R148hFxPw