Find a Question:
MIT platform learns to detect Internet attacks with expert input
Using machine learning and input from security experts manages a platform of MIT and the company PatternEx there to detect 85 percent of Internet attacks at an early stage. Moreover, the system provides few false positives.
Researchers from MIT and machine learning startup PatternEx combine four components for their system, which they have christened AI 2. It mainly concerns a platform for analyzing big data. In addition, various methods have been implemented for detection, and there is a mechanism to process feedback from security analysts present. Finally, AI 2 includes a module for supervised learning.
Initially, the platform analyzes data in search of suspicious activity using three different methods unsupervisedlearning. On the basis of large data sets brings the behavior of entities platform in card for certain periods of time, in order to be able to detect deviations. These are a list submitted to human security experts, which indicate what behavior is normal and what activities the attack as being label DDoS or data theft.
These labels are then entered as feedback with the kind of attack on the supervised learning module. This adjusts the initial analysis on the basis of the feedback, after which the system analyzes data and re-submit a new list of different activities to the expert. This process is repeated several times, which detect the system always knows better aberrant activity.
“You can think of the system as a virtual analyst,” says Kalyan Veeramachaneni. “The continuously generates new models that it can refine in a few hours, which means that it can rapidly and significantly improve the detection.” The platform security experts need to relieve the long run. Where AI 2 early two hundred different events dishes out one day, this decreases over time to thirty or forty.
The researchers have tested the platform with a dataset of 3.6 billion logs. They claim that eventually 85 percent of the tested Internet attacks was successfully detected and the number of false positives decreased by a factor of five. The researchers present their findings in a paper titled Training a big data machine to defend.Viewing:-127
Answer this Question
You must be Logged In to post an Answer.
Not a member yet? Sign Up Now »
Star Points Scale
Earn points for Asking and Answering Questions!