MIT Creates Remarkably Accurate Tool To Detect Cyber-Attacks

Posted: Apr 18 2016, 9:45pm CDT | by , Updated: Apr 19 2016, 11:34pm CDT, in News | Latest Science News


MIT Creates Remarkably Accurate Tool to Detect Cyber-Attacks
Credit: Kalyan Veeramachanen/MIT CSAIL

New system combines human intuition and machine learning and can predict cyber-attacks with 85% accuracy.

Hackers never sleep. They continue to target computer networks and damage their infrastructure. 

Now, a combined team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and machine-learning startup PatternEx have developed a powerful artificial intelligence system called AI2 which works significantly better than any existing cyber-attack detection system. 

The system has been tested on 3.6 billion log lines or pieces of data that reveal major system activities triggered by millions of users over a period of three months. Researchers have found that new tool can detect cyber-attacks with 85% accuracy which is roughly three times better than the previous benchmark. Moreover, it reduces the number of ‘false positives’ – an event wrongly identified as threat – by a factor of 5.

Conventional security systems are either virtual machine-based or humanly operated but none of them has proven overwhelmingly successful at encountering cyber-attacks. In the latest AI platform, researchers have combined them both to overcome the shortcomings associated with each of them when they work separately.

To predict attacks, system carefully looks through data and suspicious activities using supervised machine learning. Then, it reports it to human analysts to confirm which attacks are actual attacks and which are false ones and incorporates that feedback into its model for the next set of data. In this way, it continuous to improve itself and makes more reliable predictions.

“You can think about a system as a virtual analyst,” said Kalyan Veeramachaneni, one of the creators of the system from CSAIL. “It continuously generates new models that it can refine in as little as a few hours, meaning it can improve its detection rates significantly and rapidly.”

Researchers have merged human and computer based approaches and created a top notch system which demonstrates the highest level of accuracy. 

 “This paper brings together the strengths of analyst intuition and machine learning and ultimately drives down both false negatives and false positives,” said Nitesh Chawla from University of Notre Dame. “This research has the potential to become a line of defense against attacks such as fraud, service abuse and account takeover, which are major challenges faced by consumer-facing systems.”

On day one of its training, AI2 identified 200 abnormal events and presented them to experts. As it improves over time, it detected more and more actual attacks. In other words, it presented fewer and fewer data to the analyst and cut human workload.

“The more attacks the system detects, the more analyst feedback it receives, which in turn, improves the accuracy of future predictions,” said Kaylan. “That human-machine interaction creates a beautiful cascading effects.”

You May Like


The Author

Hira Bashir covers daily affairs around the world.




Leave a Comment

Share this Story

Follow Us
Follow I4U News on Twitter
Follow I4U News on Facebook

You Also Like


Read the Latest from I4U News