There used to be a maxim among technology firms that engineers used to throw a dash of cold water on energetic sales and marketing types demanding new and greater products. Sure, the engineers would say, you can have if fast (squeals of delight from marketing staff), you can have it cheap (salespeople toss hats in the air) or you can have it right. Pick two (silence).
A research team at the Flatiron Institute’s Center for Computational Astrophysics on Monday presented research on a new computer artificial intelligence (AI) simulation that generates complex three-dimensional simulations of the universe both faster and more accurately than other models. No word on price, but we can guess that it’s not exactly cheap.
The biggest surprise, however, is that the model the researchers built had never been trained on how tweaking different parameters could change the results. Shirley Ho, a group leader at the Flatiron Institute said, “We can run these simulations in a few milliseconds, while other ‘fast’ simulations take a couple of minutes. Not only that, but we’re much more accurate. … It’s like teaching image recognition software with lots of pictures of cats and dogs, but then it’s able to recognize elephants. Nobody knows how it does this, and it’s a great mystery to be solved.”
So now, in addition to unlocking the mysteries of the cosmos, Ho and her team have “an interesting playground for a machine learner to use to see why this model extrapolates so well, why it extrapolates to elephants instead of just recognizing cats and dogs.”
As the authors state it a bit more scientifically in their paper: “A striking discovery is that the Deep Density Displacement Model (D3M), trained by using a single set of cosmological parameters, generalizes to new sets of significantly different parameters, minimizing the need for training data on a diverse range of cosmological parameters.”
The neural network that powers the D3M model was trained with 8,000 different simulations and then ran simulations on a box-shaped universe 600 million light years on a side, competing with two other models running the same simulations. The slow-but-accurate approach took hundreds of hours of computation time per simulation and the existing fast method took a couple of minutes. The D3M model completed a simulation in just 30 milliseconds.
All that’s left to do now is to figure out how it all happens.