Neural Agents
Neural Agents

Detection-oriented technologies generally fall into two broad areas, signature-based detection and anomaly-based detection.    Complex event processing (CEP) is also a detection-oriented technology, so we can readily understand that CEP applications must also fall within the same two general areas.

Signature-based detection is sometime referred to as static detection because the technology relies on pre-defined rules, filters, and signatures to match known patterns.  At the most fundamental level, a virus checking program is an example of a signature-based system.

On the other hand, anomaly-based detection systems strive to maintain a baseline of what is considered normal and then matches patterns outside normal operating parameters, often usings adaptive or artifical intelligence techniques.

Experts know that both anomaly and signature-based detection methods are important and each have their unique challenges and engineering tradeoffs.  For example, signature-based systems tend to generate false negatives because it is not possible to write all possible rules and filters to match every pattern, especially in dynamic real-time environments. Anomaly-based detection, on the other hand, tends to generate false positives because it is quite difficult to create a perfect profile of normal behavior.

The challenge in most, if not all, detection-oriented systems is finding the right balance between false positives and false negatives.  In some situations, a system should error toward false positives.  In other applications, the system should error toward false negatives.

CEP is, by defination, a technology to detect both opportunities and threats in distributed networks, in real-time, so it goes without saying that CEP is challenged by the same engineering tradeoffs that affect other detection-oriented systems.

A few weeks ago, I was discussing CEP with a CTO of one of Thailand’s largest telecommunications companies and he was very bullish on neural-based anomaly detection and from Esphion.

First generation detection systems rely on determinism, which is generally rule-based, and known to be insufficient for more complex real-time problems.  Esphion uses neural agents to gathering information on network activity and then creates a unifying situational infrastructure to protect against previously unknown threats.   For example, a fast spreading threat, such as the SQL/Slammer worm, will have reached all possible targets faster than any signature can be published or rule can be written, as mentioned in Worm detection – You need to do it yourself.

Since CEP is designed and marketed as a technology that brings real-time advantages to the detection of both opportunties and threats, we must ask ourselves the question why do all the current CEP software vendors fail to provide non-deterministic methods that are proven to adapt to a rapidly changing world?

In Anomaly Detection 101, Esphion does a great job of describing how they do not rely on any pre-specified rules, baselines, models, signatures, or any other apriori knowledge.   They claim, and my highly respected telecommunications CTO colleague confirms, that there is no prior knowledge required and their customers are no longer adversely affected by zero-day anomalies or changing network conditions.

The technology behind Esphion does is what I would call complex event processing.