|March 1 · Issue #30 · View online |
| /issues/deep-learning-weekly-issue-30-most-cited-papers-black-magic-self-driving-cars-in-the-browser-and-more-46716/ |
As always, if you’d like to support deep learning weekly share this issue with friends and colleagues.
| Deploying Deep Learning Frameworks at Scale |
Algorithmia shares some insights on their architecture decisions when bringing deep learning to the platform in 2016. They go over the main challenges: Providers, languages, large models and GPU sharing, and make some predictions about model hosting.
| The Alien Style of Deep Learning Generative Design |
A look at designs generated by deep learning systems, how they differ from the normal expectations and what future applications of such designs may be.
| Self-driving cars in the browser |
| The Black Magic of Deep Learning |
Nikolas Markou drops a list of very specialized and powerful tricks to make deep neural networks work the way you want them to. He covers datasets, architecture decisions, training and more. Highly recommended!
| Feel The Kern - Generating Proportional Fonts with AI |
Patrick Gadd trains a deep neural network to generate new fonts. To do so he uses bigrams, pairs of characters, encoded as one-hot vectors and so called style vectors, that represent fonts as his networks inputs. After training he can interpolate the font style of his bigrams between existing fonts. He then added an algorithm to combine the bigrams into full words. Well written article on how to encode characters and fonts for your neural net and the source code is included.
| NIPS 2016 Workshop on Adversarial Training |
The videos of the NIPS 2016 Workshop on Adversarial Training are available on Youtube. Head right in and listen to talks from Ian Goodfellow, Yann LeCun and more.
| Generating Color Palettes with Deep Learning |
A nice article on color palette generation using GANs. It lacks most of the implementation details, but you can try the generator right on the website and get an idea on how to implement it yourself.
| Preprocessing for Machine Learning with tf.Transform |
Google enhances TensorFlow with the new tf.Transform module. This allows the definition of preprocessing pipelines that make use of large scale data processing frameworks. Those pipelines can then be exported and added to a TensorFlow graph. So now you can preprocess your data in your large-scale data environment and port this pipeline into your model for inference.
| The most cited deep learning papers |
A curated list of the top 100 deep learning papers from 2012 to 2016. The papers are weighted by the number of their citations and need to fulfill some criteria, which hopefully leads to a list of must read papers.
| Using Deep Learning and Google Street View to Estimate the Demographic Makeup of the US |
A method that determines socioeconomic trends from 50 million images of street scenes, gathered in 200 American cities by Google Street View cars. Using deep-learning-based computer vision techniques, the authors determined the make, model, and year of all motor vehicles encountered in particular neighborhoods.
| Deep Reinforcement Learning: An Overview |
An overview of recent exciting achievements of deep reinforcement learning (RL).