In brief The US Department of Energy has set aside $21m (£16m) to fund the use of machine learning to advance fusion power.
“These awards will enable fusion researchers to take advantage of recent rapid advances in artificial intelligence and machine learning,” said Chris Fall, Director of DOE’s Office of Science. “AI and ML will help us to accelerate progress in fusion and keep American scientists at the forefront of fusion research.”
Fusion reactors release copious amounts of heat that, in turn, produces steam to generate electricity. Modern designs like the tokomak use a magnetic field to confine the hot plasma, but they aren’t used on industrial scales because it requires an incredible amount of energy to kickstart nuclear fusion reactions in the first place.
Still, they’re an attractive idea since the process is safer than nuclear plants and doesn’t emit pesky greenhouse gases that lead to climate change. But fusion seems to be one of those technologies that is always a decade away.
The DOE hopes that machine learning algorithms can make fusion energy more efficient. For example, AI systems could automate some control systems and help scientists analyze data to reveal the best way to operate fusion reactors.
AI portraits of Roman emperors from ancient busts
A designer has used an online tool to create photorealistic depictions of Roman emperors from historical sources.
Daniel Voshart has modeled 54 emperors so far, including Augustus, Nero and Marcus Aurelius using various software. Artbreeder uses a generative adversarial network to create the fake, digital images.
Voshart would feed the software depictions of the ancient leaders alongside color representations that he had edited in Photoshop. The outputs of Artbreeder would be fine-tuned in Photoshop and fed back into the system in a loop, until the results were good enough.
“I would do work in Photoshop, load it into ArtBreeder, tweak it, bring it back into Photoshop, then rework it,” he told The Verge. “That resulted in the best photorealistic quality, and avoided falling down the path into the uncanny valley.” Sometimes he even took inspiration from real photographs of people too.
An Assistant Professor of Classics at USC has informed me that he might be the doppelgänger depiction of Numerian. pic.twitter.com/0Qh66pTPpB— Dan Voshart 𓀡 𓀒 𓀓 𓀢 (@dvoshart) July 29, 2020
You can see more images here.
Who’s the fastest Formula One driver? Let’s ask AWS
Amazon’s machine learning algorithms have determined that the best Formula One driver since 1983 is Ayrton Senna.
AWS collaborated with F1 to analyze lap times, and took into account things like age of the driver, the number of crashes and even weather conditions to rank top 20 drivers. Michael Schumacher comes in second, and Lewis Hamilton third.
“With machine learning, there are a number of opportunities to apply the technology to answer complex problems, and in this case, we hope to help settle age-old disputes with fans by using data to inform decisions,” said Priya Ponnapalli, a principal scientist and senior manager at Amazon’s ML Solutions Lab.
Amazon’s results just didn’t sit right with everyone, however. So much so that F1’s managing director Ross Brawn issued a statement supporting the algorithm to reporters. “There’s been one or two surprises but when you delve into it, there’s a certain amount of sense,” Reuters reported. ®