|February 23 · Issue #29 · View online |
Another exciting week in Deep Learning:
This week we have seen the deep learning cloud wars heating up with Google following Amazon and a startup entering the scene aiming to become the Heroku of deep learning, tensorflow 1.0 was released, Microsoft open sources AirSim, a framework for training and testing drones and Pinterest offers a peek at their visual discovery engine.
If you enjoy receiving this newsletter we’d appreciate it if you shared it with your friends and colleagues.
As always happy reading and hacking!
| Neural Network Learns to Select Potential Anticancer Drugs |
Scientists from Mail.Ru Group, Insilico Medicine and MIPT have for the first time applied a generative neural network to create new pharmaceutical medicines with the desired characteristics. By using Generative Adversarial Networks (GANs) developed and trained to “invent” new molecular structures, there may soon be a dramatic reduction in the time and cost of searching for substances with potential medicinal properties.
| GPUs are now available for Google Compute Engine and Cloud Machine Learning |
Google Cloud Platform is following Amazon AWS’s GPU offering by opening their beta program making available NVIDIA Tesla K80 GPUs. You can spin up NVIDIA GPU-based VMs in three GCP regions: us-east1, asia-east1 and europe-west1, using the gcloud command-line tool.
| Floyd Zero Setup Deep Learning |
YC backed startup Floyd aims to be the Heroku for Deep Learning. This exciting offering must come as a release to anyone tinkering with their own setup and battling to get everything set up correctly on AWS who can now simply run ‘floyd run –env tensorflow’ and be good to go.
| If AI Can Fix Peer Review in Science, AI Can Do Anything |
An interesting Wired article on the use of deep learning based NLP algorithms for summarizing scientific papers thus making the peer review process more efficient and objective.
| Getting Started with Deep Learning - Silicon Valley Data Science |
A great and concise overview and comparison of deep learning frameworks, tutorials and training materials.
| Attacking Machine Learning with Adversarial Examples |
Interesting article by OpenAI that shines some light on the problem of adversarial examples in AI safety. Adversarial examples are inputs to machine learning models that an attacker has intentionally designed to cause the model to make a mistake; they’re like optical illusions for machines. The post explores how adversarial examples work across different mediums and explains why securing systems against them can be difficult.
| How deep learning and AI techniques accelerate domain-driven design |
ThoughtWorks’ Steven Lowe examines the interesting intersection between Deep Learning and Domain-Driven Design. For example, Lowe applied Word2vec to process user stories and code for a software project with the goal of comparing how closely application code matches a business domain model.
| Creating Photorealistic Images with Neural Networks and a Gameboy Camera |
| Announcing TensorFlow 1.0 |
TensorFlow 1.0 was announced as part of the first annual TensorFlow Developer Summit, hosted in Mountain View.
| AirSim: Microsoft Shares Open Source System for Training Drones |
Microsoft researchers have been working on a new set of tools that developers can use to train and test robots, drones and other gadgets for operating autonomously and safely in the real world. A beta version is available on GitHub.
| TensorFlow Debugger (tfdbg) Command-Line-Interface Tutorial |
TensorFlow debugger is a new tool that promises to make debugging machine learning models (ML) in TensorFlow easier. It provides visibility into the internal structure and states of running TensorFlow graphs. The insight gained from this visibility should facilitate debugging of various types of model bugs during training and inference.
| Generative Temporal Models with Memory |
A fascinating paper exploring the general problem of modeling temporal data with long-range dependencies, wherein new observations are fully or partially predictable based on temporally-distant, past observations.
| Visual Discovery at Pinterest |
This paper gives a comprehensive overview of Pinterest’s visual discovery engine, including object detection and recommender engine.