|September 20 · Issue #7 · View online |
this weeks issue features an algorithm marketplace, an overview of neural network architectures, a new reinforcement learning tool from Twitter and much more!
If you like this issue don’t hesitate to share it with your friends and colleagues, it would be greatly appreciated.
| Need Some AI? Yeah, There’s a Marketplace for That |
A look at Academias approach of offering machine learning algorithms to companies and developers in the form of a classic marketplace. Why not just buy the neural net you always wanted?
| The Next Wave of Deep Learning Applications |
After introducing us to the next wave of deep learning hardware in last weeks issue, this article presents upcoming applications.
| The Neural Network Zoo - The Asimov Institute |
A wonderful overview of the existing neural network architectures and their features.
| Reinforcement Learning for Torch: Introducing torch-twrl |
Twitter presents a Reinforcement Learning framework with minimal dependencies.
| On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima |
Large badges in stochastic gradient descent have been shown to lead to poorer generalization, this paper demonstrates that this behavior is due to the fact that large badges tend to converge to sharp minima (strongly unique local minima), whereas methods consistently converge to flat minimizers.
| Direct Feedback Alignment Provides Learning in Deep Neural Networks |
This paper demonstrates an elegant approch to decoupling the weights used in error backpropagation to the ones used to propagate the activation forward. The paper argues that this is a step towards biologically plausible machine learning because the error signal is almost local, and no symmetric or reciprocal weights are required.
| Playing FPS Games with Deep Reinforcement Learning |
Deep reinforcement learning in 3D environments in first-person shooter games involving partially observable states.
| Facebook AI Director Yann LeCun on His Quest to Unleash Deep Learning and Make Machines Smarter |
Great interview from last year with Yann LeCun about his Deep Learning experiences: “Imagine a box with 500 million knobs, 1,000 light bulbs, and 10 million images to train it with. That’s what a typical Deep Learning system is.”
| Naveen Rao - Deploying Deep Learning at Scale - Making Machines Smarter |
Interesting talk by Nervana’s Naveen Rao on how to deploy Deep Learning at scale.
| Salesforce Einstein delivers artificial intelligence across the Salesforce platform | TechCrunch |
Salesforce is starting to bring Deep Learning and artificial intelligence to its products by leveraging the brain power they acquired over the last year.