View Feed
group-icon
Hacking and Computer Security
Community for every ethical hacker and computer security enthusiast to discuss latest and the best concepts and ideas.
370 Members
Join this group to post and comment.
Ambarish Ganesh
Ambarish Ganesh • Apr 19, 2016

MIT's AI² Cybersecurity Algorithm Can Predict 85% Threats

The need for stronger cybersecurity is ever increasing. Indian IT body NASSCOM has claimed that by 2025, this is an area that shall globally become a $350 million industry, with more and more businesses turning to secure their company system and data. Now with the conventional we either follow specified instructions given out from an expert, or rather employ machines to track abnormal activity. But the issue here is hackers can easily crack through most of such methods.

MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) in collaboration with machine-learning startup PatternX has composed an algorithm AI² with the sole aim of information security and prediction of cyber threats. The algorithm makes use of Artificial Intelligence and Analyst Intuition, and hence the name AI².

AI2 MIT PatternX

The system begins with three varying unsupervised machine learning techniques to recognize and tag potentially suspicious activity. These are then monitored by a human analyst, who can confirm or deny whether the tagged activity is truly suspicious. The human feedback is then entered into the machine learning loop, and the following cycle of machine learning detection integrates the human input as well. AI² can scan billions of log lines per day, and transforms the data minute-by-minute into 'features' that are eventually tagged normal or abnormal. The more attacks the system registers, the more human feedback it receives, thus bettering the machine learning and improving the accuracy of the predictions.

This methodology increases the efficiency of the machine learning algorithms by putting up humans as an important element in the process, and the threat detection count came to 85 percent of attacks, which is three times better than any approach attempted earlier. This new system even brings down the number of false positives by a factor of 5.

Share this content on your social channels -