Om nom nom — what are you feeding your machine learning (ML) system? NIST researchers at the crossroads of communications technology and cybersecurity are working from that question.     

At its most basic level, ML acts in the same way your digestive tract would. Good food leads to good results. Feeding it junk food (inputting bad data) leads to failure.   

So, naturally, if you expect ML to detect a cyberattack on a system but only give it bad data to learn from, you’re going to have a bad time.  

Now, we’re demonstrating the ability to create trusted, useful data for ML, which can help detect an attack on the system. And our team is using 1,000-2,000 types of data to do it.  

Specifically, the researchers are catching and collecting data on a cyberattack, the system’s behavior when it’s under attack and the system’s behavior when it’s not under attack.  

Using that information, we’re hoping to help system administrators keep their ML’s diet nutritious. For more information (and fewer biological metaphors), check out the project page.