|October 9 · Issue #99 · View online |
Welcome to a new week in deep learning!
As always happy reading and hacking. If you enjoy this newsletter please recommend us to your friends and colleagues.
See you next week!
| A Google intern's BigGAN AI makes super realistic images |
Google and DeepMind once again presented an extremely impressive GAN, capable of generating almost lifelike images at very high resolutions. The results are so fascinating, that Twitter was flooded with weird examples
the whole week.
| Opinionated and open machine learning: The nuances of using Facebook's PyTorch |
Interesting interview with PyTorch project lead Soumith Chintala on the thinking behind PyTorchs creation, and the design and usability choices made.
| iOS 12 Core ML Benchmarks |
With the A11 Bionic’s Neural Engine in last year’s iPhone X, Apple introduced its first chip for hardware-accelerated AI. This year, Apple promised huge performance increases to Core ML in their iOS…
| OpenAI 2019 Winter Fellows & Summer Interns |
| Glossary of Machine Learning Terms |
A constantly updated machine learning glossary always comes in handy and especially in machine learning, where new buzz words are born everyday, you should have bookmarked at least one.
| Machine learning in AFL Part II - It's all about percentages |
Although we can’t really recommend betting money on your own models, this is an interesting article on modelling AFL matches. The author made a net positive return, so the models seem to have some value.
| AI still fails on robust handwritten digit recognition (and how to fix it) |
While this is actually a publication, we liked the well written introduction to ongoing issues in MNIST detection and it’s a dataset everyone coming into the field will stumble upon at some point, so why not learn why it’s still hard?
| 18 Tips for Training your own Tensorflow.js Models in the Browser |
Although Tensorflow.js is still more of an exotic framework, some helpful tips on how to use it and where to look if something goes wrong, are always helpful.
| Fully Convolutional DenseNet for semantic segmentation of images implemented in TensorFlow |
A nice and handy implementation of the ‘Tiramisu’ architecture in TensorFlow.
| Learning with random learning rates |
Very interesting new approach to learning rate tuning. Instead of looking for the one learning rate, the technique uses multiple learning rates at the same time in the same network. By sampling one learning rate per feature, it reaches performance close to the performance of the optimal learning rate, without having to try multiple learning rates.
| Exascale Deep Learning for Climate Analytics |
This paper received quite a lot of attention due to the sheer amount of resources involved. The motivation is, as so often, the weather… More details may be found here