Deep Learning: An Essential Guide to Deep Learning for Beginners Who Want to Understand How Deep Neural Networks Work and Relate to Machine Learning and Artificial Intelligence by Jones Herbert

Author:Jones, Herbert [Jones, Herbert]
Language: eng
Format: epub
Published: 2018-09-27T16:00:00+00:00

Computer systems intrusion detection

In sci-fi books such as William Gibson’s Neuromancer , hackers plug into the internet via a literal plug inserted into the base of the skull and jockey the software around; intrusion detection is done by an AI called Black Ice that guards proprietary cyberspace and tries to fry the hackers’ brains. Hackers and intrusion detection exist right now in a much more prosaic manner, but neural networks promise that at least that latter part is about to become much more interesting.

“Statistical Analysis Driven Optimized Deep Learning System for Intrusion Detection” [35] investigates the rising threat of intelligent malware and hacking attacks that might jeopardize banking systems, power grids or hospital record databases. It’s not just that everything is networked, but the sheer size of such networks makes updating a nightmare and increases their attack surface, leaving them exposed to any hacker with an opportunity; it’s as easy as walking to one of these terminals connected to the intranet , the internal network, and popping in an infected USB. It’s probably how WannaCry ransomware hit 16 UK hospitals in May 2017, locking all medical files behind a paywall and threatening deletion unless $300 in ransom was paid in Bitcoin [36] .

This research paper shows a scalable, lightweight way to keep huge networks secure, regardless of whether their components are updated or not, by using neural networks that sift through a massive volume of data to anticipate intruder behavior and deny them access. We know from other scientific areas that neural networks can achieve near-human performance in cases of person recognition and 3D object reconstruction, so it’s of great interest to find a sustainable, cheap alternative to antivirus programs and outdated control access routines that cause endless grief to support staff, such as usernames and passwords.

The neural network assigned to network security does data preprocessing to remove outliers, feature extraction to find commonalities between users, and classification to distinguish between benign and malign users. In general, intrusions come as: probing that scouts the target network for weaknesses and open ports; denial-of-service, which serves to incapacitate target network and estimate its capabilities; user-to-root that is meant to gain root access; and root-to-local that is meant to perform operations on a local machine after root access has been gained.

Neural network was first trained using 125,973 samples belonging to any of the four intrusion categories and tested with another 22,544 samples, resulting in an accuracy of 77.13% for probing attacks, 97.08% for denial-of-service, 87.10% for user-to-root but only 11.74% for root-to-local. These numbers imply that all security measures should focus on preventing access to root systems, which are those that can issue commands to subordinate machines, including the entire network or individual terminals; once the root has been hijacked, every compromised machine, or even the entire network, is decisively in the hands of the attacker, as seen with WannaCry ransomware.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.