The neural network is trained on 15,000 signals from the Kepler dataset that have been previously verified as planets or non-planets. A smaller test set with new, unseen data was fed to the neural network and it correctly identified true planets from false positives to an accuracy of about 96 per cent.The researchers then applied this model to weaker signals from 670 star systems, where scientists had already found multiple known planets to try and find any that might have been missed.Vanderburg said the got lots of false positives of planets, but also more potential real ones too. “It’s like sifting through rocks to find jewels. If you have a finer sieve then you will catch more rocks but you might catch more jewels, as well,” he said.
Robin Edgar
Organisational Structures | Technology and Science | Military, IT and Lifestyle consultancy | Social, Broadcast & Cross Media | Flying aircraft