Fluid & Smoke Simulation with Deep Learning, Evolutionary Algorithms, a Self-Driving Car From the 80s & More
Transmission #4
This week’s newsletter includes a self-driving car from 1989, news on Amazon’s deep learning efforts, real-time simulation of fluid and smoke using deep learning, image-to-image translation, a new deep learning supercomputer in Japan and more. Enjoy!
Each week I (@olivercameron) will be sharing the very latest news in deep learning and self-driving cars. To get priority access to these newsletters, please join the mailing list at transmission.ai!
Meet ALVINN, the Self-Driving Car From 1989
Learn more about ALVINN (Autonomous Land Vehicle In a Neural Network), a groundbreaking project from CMU in the late 80s to build an autonomous vehicle powered by a neural network. Read more…
Amazon Adopts MXNet for Deep Learning
Amazon has adopted MXNet as their deep learning framework of choice for AWS customers, signaling where their focus will be for any deep learning efforts (vs. TensorFlow etc.). Read this note from Werner Vogels (Amazon CTO) for more information…
Google & NYU Open Sources FluidNet
Real-time simulation of fluid and smoke is a tough problem. This paper proposes using deep learning to obtain both fast and highly realistic simulations. Crazy! Read more and check out the code…
Image-to-Image Translation
An exciting paper (with code) on using conditional adversarial networks to predict output images from an input. For example: given a Google Maps tile, output an image of the predicted aerial view. Very cool! Read more…
Japan Plans Deep Learning Supercomputer
Japan aims to build a 130 petaflop machine that will be accessible to others (for a fee), that appears to be focussed on deep learning applications. The closest rival is a 93 petaflop machine in China, so this would be a significant leap forward if built. Dubbed ABCI (AI Bridging Cloud Infrastructure), the project will cost around $173 million. Read more…
End-to-End English Speech Recognition Using DeepMind’s WaveNet
DeepMind’s WaveNet paper was a revelation to many, and now you can play around with a TensorFlow implementation! Although not a 1:1 replica (the paper has gaps on implementation details), it’s a good way to get started learning. Read more…
Evolutionary Algorithms Could Be More Significant Than Machine Learning
An interesting look at how evolutionary algorithms might evolve (pun intended) to be more significant than machine learning. My take: the best tool for the use-case should win! Read more…
An Interactive Tutorial on Numerical Optimization
A fantastic and interactive way to learn more about numerical optimization, a core technique used in machine learning. We need more tutorials like this! Try it out…
Uber Open Sources deck.gl
Uber has open sourced deck.gl, a WebGL-powered framework for visually exploring large datasets. Take look at a demo of the mapping integration and prepare to be wow’d! Read more…
That’s it for this week, thanks for reading! If you have any thoughts or questions, I’d love to hear from you in Tweet-form. You can follow and message me at @olivercameron.
