HOW INTERCEPT X USES DEEP LEARNING TO FIGHT SECURITY THREATS

Intercept X Deep Learning
 

Intercept X combines deep learning with best-in-class anti-exploit technology, CryptoGuard anti-ransomware, root cause analysis, and more to form the industry’s most comprehensive endpoint protection. This unique combination of features allows Intercept X to stop the widest range of endpoint threats.

Deep Learning vs. Other Types of Machine Learning

Intercept X uses a deep learning neural network that works like the human brain… This results in a high accuracy rate for both existing and zeroday malware, and a lower false positive rate.

ESG Lab ReportDecember 2017

Much of today’s security is reactive and far too slow. As the volume and complexity of endpoint attacks has continued to grow, legacy approaches have struggled to keep pace. For example, SophosLabs analyzes over 400,000 new malware samples every day. To make meeting this challenge even more difficult, SophosLabs found that 75% percent of malware is unique to a single organization.

 

Deep learning, an advanced form of machine learning, is helping to change the way we approach endpoint security, and Intercept X is leading the charge. By integrating deep learning, Intercept X is changing endpoint security from a reactive to a predictive approach to protect against unknown threats.

 

While many products claim to use machine learning, not all machine learning is
created equally. At Sophos, we use deep learning to detect malware. Also referred to as ‘deep learning neural networks’ or ‘neural networks’, deep learning was inspired by the way the human brain works. It is the same type of machine learning often used for facial recognition, natural language processing, self-driving cars, and other advanced fields of computer science and research.

 

Deep learning has consistently outperformed other machine learning models, including random forest, k-means clustering, or Bayesian networks, but requires vast amounts of data and computational power to build an effective model. At Sophos, this has been made simple thanks to the malware collection and analysis efforts of SophosLabs over the past 30 years and the telemetry we receive from our 100+ million endpoints every single day.

Highlights

  • The number one performing malware detection engine
  • Prevents both known and never-seen-before malware
  • Blocks malware before it executes
  • Does not rely on signatures
  • Protects even when the host is offline
  • Detects malware in approximately 20 milliseconds
  • Trained on hundreds of millions of samples
  • Proven on VirusTotal since August of 2016
  • Classifies files as malicious, potentially unwanted apps (PUA), or benign
  • Works out-of-the-box with no additional training needed
  • Extremely small footprint (under 20MB)
  • Focused on Windows portable executables
Intercept X stopped every complex, advanced attack we threw at it

ESG Lab ReportDecember 2017

Smarter

 

Deep learning models process data through multiple analysis layers, just like neurons in the human brain, each layer making the model considerably more powerful. It analyzes complex relationships between different input features. This allows it to automatically uncover the best combination and manipulation of inputs that would otherwise be impossible for humans to determine. This means that the Sophos deep learning malware detection model will be able to detect malware that would go unnoticed by other machine learning engines.

More Scalable

 

Deep Learning elegantly scales to hundreds of millions of training samples. This is important considering that SophosLabs analyzes 2.8 million new malware samples every week. Because it can continue to ingest massive amounts of training data our model can ‘memorize’ the entire observable threat landscape as part of its training process. Since it can process significantly more input, deep learning can more accurately predict threats today while continuing to stay up-to-date over time.

Lighter

 

Traditional machine learning approaches result in huge model sizes, which can sometimes take many gigabytes on disk. However, Sophos’ deep learning approach results in highly compressed models. The Sophos deep learning model is incredibly small, less than 20MB on the endpoint, with almost zero impact on performance.

Experienced

 

Unlike the competition, we have been cybersecurity machine learning experts for a long time, and have had our malware detection deep learning models in production environments for years. The Sophos malware detection model was created by our team of data scientists with DARPA driven technology. In 2010, the US Defense Advanced Research Projects Agency (DARPA), created their Cyber Genome Program to uncover the ‘DNA’ of malware and other cyber threats. This was the origin of what is now the algorithm embedded in Intercept X.

Proven

 

We have been open and transparent with our models. In addition to presenting details of our methodology at industry conferences such as Black Hat, we also have not shied away from allowing our model to be tested by independent third parties. The model has been proven on VirusTotal since August of 2016, and has received high scores from third-party testers such as NSS Labs. In all cases, it has proven to be extremely effective while having low false positives.

One of the best performance scores we have ever seen in our tests

Maik Morgenstern, CTOAV-TEST

Performance

 

Sophos’ Deep Learning technology is incredibly fast. In less than 20 milliseconds the model is able to extract millions of features from a file, conduct a deep analysis, and determine if a file is benign or malicious. This entire process happens before the file executes.