This week we look at the 40% drop in car crashes using Tesla's Autopilot, why Microsoft is investing into Montreal AI research, check out some algorithms IRL, and building neural networks in Python.
Not a subscriber? Sign-up to Emergent // Future here.
Last week: we looked at training self-driving cars using Grand Theft Auto V and the CB Insights AI 100 landscape.
Want to contribute to Emergent // Future? Reply and let me know!
Tesla Autopilot > Human-Driven Cars
The U.S. Department of Transportation closed it’s probe into Tesla’s Autopilot after data showed a 40% drop in car crashes when the feature is enabled.
Tesla was able to provide the impact of Autopilot on crash rates, since not all Tesla owners have enabled the self-driving software.
The data showed a 1.3% crash rate per million miles driven without Autopilot enabled vs a .8% crash rate with.
OH? Postmates and DoorDash are testing self-driving delivery bots in California and DC.
Montreal’s AI Moment
Microsoft is doubling the size of their AI research outpost in Montreal after acquiring the Canadian startup, Maluuba, earlier this month.
Microsoft will invest $7M Montreal-based AI research at the University of Montreal and McGill University over the next five years, in addition to doubling the Maluuba team over the next two years.
Google invested invested $3.4M in the Montreal Institute for Learning Algorithms in late 2016. Google also plans to open an AI research group in their Montreal offices.
Here’s three examples of algorithms in the wild:
- The CMU poker AI, Libratus, has built a substantial in the Brains vs AI competition. Libratus has amassed a lead of almost $500,000 in chips in 49,240 hands of poker over the past nine days. The 20-day, 120,000 hand contest is a heads-up, no-limit Texas Hold’em poker tournament where four of the best poker players take on Libratus. [quietly withdraws cash from Bovada account]
- Video of the world’s first Fishstronaut. Watch this guppy pilots his own fishtank.
- Two Copenhagen-based designers ran The Wolf of Wall Street trailer through a neural network (YOLO-2) and captured the real-time object detection. Watch the algorithm try to make sense of things here.
- And, here’s some Jedi mind tricks: an interactive brain-computer interface. Artist and researcher Alessio Chierico uses an EEG device to detect brain waves, attention, and meditation levels in users gazing at an open flame. The attention levels corresponded a flow of air under the flame. The more the user concentrates on the flame, the stronger the flow of air. By focuing their attention on the flame, the user is able to extininguish the flame using their mind. Poof.
Neural Networks in Python
Facebook’s AI team released a PyTorch, a Python implementation of the Torch machine learning framework.
The package enables GPU-accelerated deep neural networks in Python, giving developers the ability to arbitrarily change the way their neural nets behave without having to build the network from scratch.
Developers can also extend PyTorch with popular Python packages, like NumPy and SciPy.
What We're Reading
- Engineering is the bottleneck in (Deep Learning) Research. There are two issues that stand out to me, both of which can be solved with “just engineering.” 1. Waste of research time and 2. Lack of rigor and reproducibility. (Denny Britz)
- Machine Learning as a Service (MLaaS) with Sklearn and Algorithmia. Separation of ML concerns from the rest of your app: By calling and making predictions as an API call, your app can be free of the concerns of machine learning mechanics. (Nick Rose)
- Inside IMAX’s Big Bet to Rule the Future of VR. The company that turns cinema to spectacle is betting big on the future of face-computers. It is the future of the movie theater—even the future of movies. (WIRED)
- Your Clothes Will Be on the Radio. In the early 2000s, it seemed as if radio frequency identification -- RFID, for short -- was about to take over the world. The tag on your brand-new sweater might hold retail's answer to Amazon. (Bloomberg)
- A curated list of software engineering blogs. #meta. (GitHub)
- Emergence. The story of natural laws and processes, their inherent beauty, and their action to yield the universe, us and the world we live in. (Max Cooper)
Things To Try At Home
- A CNN that identifies water in satellite images
- Generic Foreground Segmentation in Images
- Applying deep learning to Related Pinterest Pins
- Learn TensorFlow and deep learning, without a Ph.D
- Raspberry Pi controlled RC car using the autopilot
- Data and Code Behind the Stories and Interactives at ‘FiveThirtyEight'
Emergent // Future is a weekly, hand-curated dispatch exploring technology through the lens of artificial intelligence, data science, and the shape of things to come. Subscribe here.
Lovingly curated for you by Algorithmia
Follow @EmergentFuture for more frontier technology