|January 25 · Issue #25 · View online |
Hello deep learning practitioners, enthusiasts, dabblers and dilettantes,
We also want to thank everyone who shared and promoted our issue last week, your support is very much appreciated! 💪
| Deep Learning AI for NASA Powers Earth Robots |
Originally tasked with developing AI software for hardware constrained systems such as the Mars Rover, Deep Learning company Neurala sets its sights back on more earthly concerns by building powerful AI systems that run on smartphone chips to power robots, drones, and self-driving cars and snatch up $14 million in funding in the process
| Microsoft to double its Montreal AI R&D office and invest $7M in academic research |
Following its acquisition of Montreal-based Maluuba Microsoft is ramping up its investment in this burgeoning AI hub.
| First FDA Approval For Clinical Cloud-Based Deep Learning In Healthcare |
The FDA approved the application of cloud-based deep learning services provided by the medical imaging provider Artery
. The service is designed to help physicians understand how a heart is functioning through deep learning powered medical imaging.
| DeepTraffic | A gamified simulation of typical highway traffic |
A fun, interactive traffic simulation that lets you train a network that performs well on high traffic roads.
| MIT 6.S094 | Introduction to Deep Learning and Self-Driving Cars |
More self-driving cars! Following Udacity’s wildly popular course on autonomous vehicles, MIT launches “6.S094: Deep Learning for Self-Driving Cars”.
| Deep Learning Nanodegree Foundation | Udacity |
Siraj Raval, the Bill Nye of machine learning, teams up with Udacity to put together an introduction to Deep Learning that promises to be in equal parts informative and entertaining.
| Engineering is the bottleneck in (Deep Learning) Research – Denny's Blog |
Insightful post by Denny Britz on why it’s essential to deep learning research to stipulate engineering standards that make results easily reproducible and comparable. Required reading for every machine learning researcher.
| Understanding the new Google Translate |
A very well-presented post explaining the inner workings of Google Neural Machine Translation (GNMT) with lots of visualizations. When you’re done with the post, take a stab at the original research paper
| PyTorch | Tensors and Dynamic neural networks in Python with strong GPU acceleration |
PyTorch is the latest addition to the growing landscape of Deep Learning libraries. It offers a simple imperative interface and dynamic graph construction. Get started with this introductory tutorial
| Domain Transfer Network | Tensorflow implementation of unsupervised cross-domain image generation |
| Kur | Descriptive Deep Learning |
The folks over at Deepgram
a company that applies Deep Learning to audio data released Kur a system for quickly building and applying state-of-the-art deep learning models. Interestingly, Kur works with specification files so that you can train and evaluate models without needing to write a single line of code.
| Attention Transfer | Improving Convolutional Networks |
Example: attention transfer for facial recognition
| Bringing Impressionism to Life with Neural Style Transfer in Come Swim |
You know the hype cycle has peaked when Hollywood stars are publishing deep learning papers. To be fair, this paper does demonstrate an interesting application of Neural Style Transfer to the big screen, in this case, Kristen Stewart’s ‘Come Swim’.
| Towards Principled Methods for Training Generative Adversarial Networks |
This paper sets out to make theoretical steps towards fully understanding the training dynamics of generative adversarial networks. Specifically, they explore common problems that arise when training generative adversarial networks such as instability and saturation.