|February 8 · Issue #27 · View online |
| Inside Libratus, the Poker AI That Out-Bluffed the Best Humans |
Libratus, an artificial intelligence developed by Carnegie Mellon University, made history by defeating four of the world’s best professional poker players in a marathon 20-day poker competition.
| How Google's Amazing AI Start-Up 'DeepMind' Is Making Our World A Smarter Place |
Google’s DeepMind has its fingers in a whole lot of pies whether reducing the power consumption of Google’s data center, beating world champions at Go or Arcade games, or streamlining radiology treatment plans, no other company has demonstrated the amazing and revolutionary capabilities of Deep Learning and contributed to these interesting times we live in.
| Google Research: Advancing Research on Video Understanding with the YouTube-BoundingBoxes Dataset |
Google released a massive dataset of 5 million bounding boxes spanning 23 object categories, densely labeling segments from 210,000 YouTube videos. This is a unique and promising dataset for training object detectors making it possible to reason about scene layout, persistence, and occlusions,
| Google Research: Using Machine Learning to predict parking difficulty |
On the other end of the spectrum, Google Maps now includes a handy, new feature which predicts parking difficulty close to your destination across 25 US cities.
| Oxford Deep NLP 2017 Course |
Lecture slides and course material for Oxford Deep NLP course taught by lecturers from DeepMind and NVIDIA.
| Demystifying Word2Vec |
We put out a little write-up this week on different ways to derive word embeddings and how they contrast with Bag of Words methods.
| Use your eyes and Deep Learning to command your computer |
A fun tutorial on how to use Recurrent Neural Networks to detect eye movements and then use them to trigger shortcuts on a computer.
| GitHub - PaintsChainer: Line Drawing Colorization using Chainer |
PaintsChainer - Use CNNs to colorize your sketches automatically.
| GitHub - WassersteinGAN |
| Understanding Deep Learning Requires Rethinking Generalization |
A fascinating paper that explores the fact that traditional approaches such restricting the model family or regularization fail to explain why large neural networks generalize well in practice.
Incidentally, this paper had the highest reviewer score of all ICLR 2017 submission:
| Pixel Recursive Super Resolution |
The authors present a model that synthesizes realistic details into images while enhancing their resolution. It does so by using a PixelCNN to define a strong prior over natural images and jointly optimizes this prior with a deep conditioning convolutional network.
| Understanding trained CNNs by Indexing Neuron Selectivity |
An interesting approach to explain the inner workings of CNNs by describing the the activity of individual neurons by quantifying their inherent selectivity to specific properties.
| Towards "AlphaChem": Chemical Synthesis Planning with Tree Search and Deep Neural Network Policies |
This paper makes use of a similar algorithm used by DeepMind’s (Monte Carlo Tree Search and Deep Policy Gradients) to plan the chemical synthesis of organic molecules, for example drugs, agro- and fine chemicals.
| Deep Reinforcement Learning: An Overview |
A succinct overview over recent deep reinforcement learning advances.