The use of artificial intelligence in security systems provides more flexibility, especially with new cyber threats always emerging. Namely, machine learning has garnered much attention for its involvement and improvement of security systems.
Most people use the term “artificial intelligence” loosely these days, but it traditionally refers to the theory and development of computer systems that may perform human tasks. Machine learning is a type of AI that allows a computer to learn, grow, and change when presented with new data.
The evolution of AI can be best described in three stages. First is the basic expert system. If we used this system to help distinguish between a dog and a cat, for example, it would use a single feature such as number of teeth to make the decision. Second is the probability-based system, which evaluates different factors (ex. number of teeth, weight, size) to determine the probability (expressed as a percentage) of the object being a cat or dog. Lastly is deep learning, which uses seemingly endless amounts of labeled samples to differentiate between cats and dogs.
If we applied these to antivirus systems, you could understand how a basic expert system would be weak and need constant updating for new threats. The probability-based system would be a bit stronger, however, only so many features would prove relevant resulting in disregarded data. Deep learning seems the most promising, and a startup called Deep Instinct is looking to develop this approach for cyber security.
Within the Deep Instinct laboratory, the deep learning system is trained on all the known samples of malware, which takes about a day to complete. The process requires heavy-duty graphical processing units to analyze the data, and the end result is a trained system about a gigabyte in size. It is then reduced to about 20 megabytes and can be installed on any endpoint device (including mobile). It works to analyze any incoming threats within a few milliseconds to keep your devices safe.
To keep the system up -to-date, new malware samples are added every few months, and updates are automatically sent to the end point devices. But even if the system is not updated for months, the small brains within the end point devices remain vigilant and can detect new files. The success rate is promising and deep learning systems will likely gain more popularity over time.
While deep learning systems are great for detecting threats, they are not the best for explaining how they did it. Eureqa is a proprietary AI engine from Nutonian whose main job is to find out why things happen. It has proven very valuable for researchers and journal publications, but it also plays a role in cyber security by helping to determine the anatomy of a cyber attack.
Still, cyber security can be a tricky mess. Constant updates are necessary thanks to appearance of new threats and attacks daily. Even though you are employing security systems to protect your data, there are still vulnerabilities between updates. And during that time, hackers can use the security software to test their attacks until something breaks through, leaving numerous customers at risk.
Tailoring your cyber security approaches can help to combat this. For example, Masergy Communications is a managed networking company which uses a combination of both local and global factors to predict and prevent cyber security issues or attacks. The unique local indicators help to improve accuracy.
Acuity Solutions offers the BluVector appliance which uses machine learning for cyber threats, and also uses a local and global approach. The pre-trained engine learns what a benign code looks like, receives updates based on global data, but also engages in new learning based on the individual customer. While the global data is shared, the customer-specific data is not, creating a more unique and secure experience.
As discussed, artificial intelligence and machine learning can greatly benefit different aspects of cyber security. What are your predictions on the future of AI in security solutions? Share with us on Facebook, Google+, Twitter, LinkedIn, and Pinterest!
For a wide range of security cameras and surveillance equipment, please visit SecurityCamExpert.com. To speak with a representative or request a site survey, please call 1-888-203-6294.
Cyber security is an ongoing problem, especially as our technology becomes a larger part of our everyday lives. Neither our efforts alone nor artificial intelligence (AI) have achieved a fool proof method. Our human-driven techniques are based on rules set by living experts, which means we miss any threats that don’t follow these rules. Machine-learning approaches, on the other hand, rely on anomaly detection, which often trigger “false positives” (nonthreats mistakenly identified as threats) that need to be checked by humans.
What if humans joined forces with AI? Researchers at MIT have been working on this concept and have produced some promising results.
AI² is a collaboration between researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the startup company PatternEx. After three months of testing on 3.6 billion pieces of user-made data, it is said to detect 85% of attacks and reduce the number of false positives by a factor of five.
To find the most important potential problems, AI² uses machine learning first, and then shows the top events to analysts for labeling. After the analysts go through and confirm which events are actual attacks, the system incorporates this feedback into its models for the next set of data. This ongoing cycle of improvement decreases potential events and increases accuracy.
What are your thoughts on AI² and the interaction of humans and AI for cyber security measures? Share with us on Facebook, Google+, Twitter, and Pinterest. You can also find us on Vine.
Visit SecurityCamExpert.com for the best selection and prices on quality security cameras and CCTV surveillance equipment. If you have any questions and would like to speak with us directly, please call 1-888-203-6294.