|August 18 · Issue #53 · View online |
Hi and welcome to another issue of Deep Learning Weekly!
Happy hacking and reading. As always if you enjoy receiving this newsletter you can help us by sharing it with friends and colleagues.
See you next week!
| AI Is Taking Over the Cloud |
Cloud storage company Box just announced, that it will enable more sophisticated search functionality by incorporating Googles Cloud Vision API. This allows search for contents of an image, instead of plain filename matching and demonstrates the abilities of machine learning in the cloud. At the same time it tells a lot about the influence of the major machine learning players, as Box uses Google’s off the shelf solution, instead of something they built on their own.
| Did Elon Musk’s AI champ destroy humans at video games? It’s complicated |
Last weeks OpenAI bot vs. human Dota 2 match has created a very vivid discussion, about hype, ‘cheating’ and what the match actually means for AI. The Verge gives an overview and joins WildML i
n saying that the challenge was significantly easier than the famous Go siege of AlphaGo.
| Automatic image retouching on your phone |
Google and MIT teamed up to boost the camera performance in mobile phones using a newly developed machine-learning photography system. The system works so fast, that it’s able to post process the images in real-time and the phone can show the processed image to the user immediately. For more details, take a look at the corresponding paper
| Google Hires Former Star Apple Engineer for Its AI Team |
Chris Lattner, known for his work on Apples Swift programming language, has joined the Google Brain Team after a short run at Teslas AI department.
| dformoso/deeplearning-mindmap |
A pretty helpful mindmap that summarizes deep learning concepts and may help you get started when you want to tackle a specific problem. For a more global view, check out the machine learning mind map
| Cooperatively Learning Human Values |
This article describes an issue, known as the ‘Value Alignment Problem’, which leads to undesired behavior when training models using reinforcement learning. This problem is the problem of matching AI objectives to ours. The article describes the problem in detail and shares solutions on how to identify and fix the issue.
| Concepts of Advanced Deep Learning Architectures |
A nice little summary of common deep learning architectures and the underlying concepts and their positive and negative aspects.
| Graph Convolutional Networks |
Thomas Kipf has created a brief introduction to graph convolutional networks and explores their applications and focuses on two papers from last year.
| TensorFlow 1.3.0 |
TensorFlow has reached a new version and offers some new estimators, as well as many different bug fixes and optimizations.
| TVM: An End to End IR Stack for Deploying the Deep Learning Workloads to Hardwares |
This new framwork is supposed to allow researchers and practitioners in industry and academia to quickly and easily deploy deep learning applications on a wide range of systems, including mobile phones, embedded devices, and low-power specialized chips — and do so without sacrificing battery power or speed.
| robbiebarrat/art-DCGAN |
Want to create art using GANs? Robbie Barrat has got you covered with this DCGAN implementation.
| Superresolution with semantic guide |
Mike Tyka shares details on his work on high-resolution GAN faces. He managed to achieve resolutions of up to 768x768 pixels and more by incorporating tiling and semantic guidance.
| Direct-Manipulation Visualization of Deep Networks |
Google has shared some details on how they built TensorFlow playgrounds
and what it means to visualize a neural network.