Canadian AI, Momentum, Big Data lanscape, Differentiable Neural Computer, Federated Learning, a Sentiment Neuron, and much more
ARTIFICIAL INTELLIGENCE NEWS #59 April 20th 2017
In the News
"Bringing A.I. home is a priority for the Canadian government, companies, universities and technologists. The goal, they say, is to build a business environment around the country’s expertise and to keep the experts its universities create in the country."
No one really knows how the most advanced algorithms do what they do. That could be a problem.
Great review of the Big Data landscape by Matt Turck of Firstmark Capital, along with a comprehensive map of all the big players.
We often think of Momentum as a means of dampening oscillations and speeding up the iterations, leading to faster convergence. But it has other interesting behavior. It allows a larger range of step-sizes to be used, and creates its own oscillations. This excellent article will give you more details and let you play with momentum through interactive visualizations.
"We used computer vision and deep learning advances such as bi-directional Long Short Term Memory (LSTMs), Connectionist Temporal Classification (CTC), convolutional neural nets (CNNs), and more."
Looking at how the popularity of Deep Learning frameworks, models and optimization algorithms has evolved.
Google shares a few details on its "Collaborative Machine Learning" approach where models are trained in a de-centralized manner. Could be used to collaboratively train models directly on users' smartphones without having to upload personal data.
Chatbots have never been able to empathize. That looks set to change, thanks to a Chinese team that has built a chatbot capable of conveying specific emotions.
Software tools & code
Interesting introduction to the way white blood cells are currently counted and how Deep Learning could help. With photos, code and data.
DeepMind announced last year that it had created a Differentiable Neural Computer (a sort of memory-augmented neural network). They have now open-sourced an implementation.
OpenAI has developed an unsupervised system which learns an "excellent representation" of sentiment, despite being trained only to predict the next character in the text of Amazon reviews.
DeepMind's open-source library for constructing neural networks on top of Tensorflow. The library uses an object-oriented approach, similar to Torch/NN, allowing modules to be created which define the forward pass of some computation.
"Lightweight, modular, and scalable Deep Learning framework" by Facebook.
Google recently published a paper about the performance of its Tensor Processing Unit (TPU) and how it compared to Nvidia’s Kepler-based K80 GPU working in conjunction with Intel’s Haswell CPU. The TPU's deep learning results were impressive compared to the GPUs and CPUs, but Nvidia said it can top Google's TPU with some of its latest inference chips, such as the Tesla P40.
Nvidia unveiled its successor to the Titan X. With 12GB memory, 3,840 CUDA cores and 12 TFLOPs
This newsletter is a collection of AI news and resources curated by @dlissmyr. If you find it worthwhile, please forward to your friends and colleagues, or share on your favorite network!
Share on Twitter · Share on Linkedin · Share on Google+
Suggestions or comments are more than welcome, just reply to this email.