|January 3 · Issue #22 · View online |
Hiya and a Happy New Year!
Enjoy your read!
As always we appreciate your support by sharing this newsletter with your friends and colleagues and would love to see you joining our deep learning study group here
| Deep Learning 2016: The Year in Review |
We sat down and took a look at all the developments that took place in 2016. The article covers major changes in infrastructure, tools, datasets and algorithms and we talk about upcoming trends in the industry.
| We chat with deep learning company, Skymind, about the future of AI |
This interview with Chris Nicholson from Skymind gives nice insights about the future of deep learning and sheds some light on real world applications, upcoming sectors and the usual AI world domination.
| DeepMind’s work in 2016: a round-up |
The DeepMind founders take a look back at the year, including their publication of two Nature papers, the success of AlphaGo, and the first signs DeepMinds of real-world impact.
| A Guide to Deep Learning by YerevaNN |
A well written, clearly structured guide covering basics, different architectures and future developments. Highly recommended!
| Generative Adversarial Networks are the hotness at NIPS 2016 |
A brief introduction to GANs, featuring a toy example and an IPython notebook.
| Summary of NIPS 2016 |
Eric Jang gives an extensive summary of this years NIPS conference and shares his thoughts on the conference.
| GANs will change the world |
More GANs, but this time more focused on the application of these networks. Includes interesting ideas and brave predictions on the upcoming role of GANs.
| Recurrent Neural Network Tutorial for Artists |
| Deep Learning frameworks: a review before finishing 2016 |
Nice overview on the current state of deep learning frameworks. Covers the major players like TensorFlow and caffe and newcomers like DL4J and mxnet.
| FastMask: Segment Multi-scale Object Candidates in One Shot |
Hu et al. managed to improve Facebooks DeepMask by expanding the body/head architecture with a head. This leads to better and especially faster results, allowing almost real-time segmentations.
| YOLO9000: Better, Faster, Stronger |
Impressive real-time object detection on 9000 object categories combined with a great name.
| iSee: Using deep learning to remove eyeglasses from faces |
Melissa Runfeldt tried to fix the problem of choosing new glasses while wearing the old ones. She describes her thoughts and process in a nice way and gives useful insights on preprocessing and hyperparater tuning.
| The Predictron: End-To-End Learning and Planning |
Silver et al. present an abstract model for planning processes called the predictron architecture thats trained end-to-end and yields better results than existing approaches.