Thu Feb 20 2020
But don’t worry – They’re not due to arrive for hundreds of years (if at all)!
Why is cloudThing talking about astrophysics and asteroids today?
Well apart from the fact that we’re all self-confessed geeks who perk up at any mention of anything outer space, this particular prediction was made by an Artificial Intelligence (AI) program and that is right up our ‘digital’ street!
The neural network program that made the prediction was developed by a team of researchers at Leiden University in the Netherlands.
Called the Hazardous Object Identifier (we’d have called it asteroidThing but no one asked us to help naming it) it’s identified eleven possible asteroids that could come close to hitting Earth.
Well… we say close…
In astronomical terms they’re predicted to come within 0.5 astronomical miles which works out as about 4.7 million miles or 7.5 million kilometres… so not that close.
They’re also not predicted to get here till between 2131 and 2923 so don’t set an alarm just yet.
These asteroids had been previously identified by NASA as non-hazardous but those new figures generated by the AI software would be enough to make it onto their database watchlist of potentially hazardous objects.
So now we know we’re safe and don’t require saving by Bruce Willis and a plucky band of Space Miners… The Netherlands research team fed in hundreds of thousands of simulated asteroid collisions with the Earth into their AI software by recoding the motion of said asteroids over time, using the ALICE supercomputer at their university.
The data on each asteroid’s trajectory could then be analysed by the machine-learning software so it could start to identify common features and patterns within the data-sets that could possibly identify how and when an asteroid might or might not hit the Earth.
If you rewind the clock, you will see the well-known asteroids land again on Earth.
This way you can make a library of the orbits of asteroids that landed on Earth.
The team at Leiden University also added data from hundreds of thousands of other, real asteroids, taken from NASA’s dastcom5 dataset.
The team then performed a ‘blind test’ on the AI by using real data they had previously kept from the software.
They tested it on data concerning nearly 2,000 asteroids previously identified by NASA as potentially hazardous, with their AI software able to correctly identify 90.99% of them as hazardous or non-hazardous.
The Dutch team commented in their recent paper that these eleven asteroids had probably been classified as non-hazardous previously due to their orbits being so erratic, making accurate predictions of their trajectory more difficult.
However this still seems like another win for Artificial Intelligence and machine learning, once again proving that the more data that can be fed in, the more accurate it’s analysis’s can become.
Interested in how Data Science, Machine Learning and Artificial Intelligence can transform your business and disrupt your sector? Why not have a chat with one of our Data Scientists?
Thu Feb 20 2020