Google’s machine learning algorithm gets human help in quest for fusion power

New Atlas Rich Haridy | July 25, 2017

Google has been using its machine learning algorithms to help speed up progress in research that looks to produce electricity from nuclear fusion (Credit: Tri Alpha Energy / Erik Lucero)

Hot on the heels of last month’s nuclear fusion breakthrough comes the first results from a multi-year partnership between Google and Tri Alpha Energy, the world’s largest private fusion company. The two organizations joined forces in 2014 in the hopes that Google’s machine learning algorithms could advance plasma research and bring us closer to the dream of fusion power.

The challenge Tri Alpha Energy faced was that the enormous experimental complexity of its plasma research involved so many variables that it was desperately in need of some advanced computing networks to help wade through the data. But it turned out that even Google didn’t have the computational resources to easily address this problem.

“The reality is much more complicated,” explains Ted Baltz, from Google’s Accelerated Science Team, “as the ion temperature is three times larger than the electron temperature, so the plasma is far out of thermal equilibrium, also, the fluid approximation is totally invalid, so you have to track at least some of the trillion+ individual particles, so the whole thing is beyond what we know how to do even with Google-scale computer resources.”

The team developed a unique solution dubbed the “Optometrist Algorithm” that fused man and machine. Deriving its name from the process of an eye exam, the algorithm presents human experts with successive pairs of possible outcomes and lets those experts use their judgment to choose between the two to direct subsequent experiments.

“So we boil the problem down to let’s find plasma behaviors that an expert human plasma physicist thinks are interesting, and let’s not break the machine when we’re doing it,” says Baltz.

The incorporation of this technique into TAE’s experimental processes allowed research to progress at an incredibly fast rate. A new study published in the journal Scientific Reports shows the algorithm unexpectedly netting the team a 50 percent reduction in energy loss rate and a concomitant increase in ion temperature and total plasma energy in TAE’s field-reversed configuration plasma generator.

“Results like this might take years to solve without the power of advanced computation to rapidly scale our understanding of the complex properties of plasma,” says Michl Binderbauer, TAE’s President and Chief Technology Officer.

Since completing this research TAE has built a new, larger plasma generator. Named “Norman”, after the company’s late co-founder Norman Rostoker, this is the fifth generation of the machine that uses an advanced field-reverse configuration (FRC) combined with intense neutral beam injection to create and confine plasma.

Google and TAE will continue to collaborate on experiments with this new plasma generator and Binderbauer suggests they could be producing electricity from fusion power within a decade.

Source: Tri Alpha Energy, Google Research Blog