What is an “Analyzer”?
A Primer on ThreatEye’s* Approach to Machine Learning
At CounterFlow AI, the term “analyzer” describes the programs or scripts that fuel its intelligent sensor. Up until now, the vernacular of machine learning has centered around “models”. An analyzer, however, is a broader term that encompasses the entire scriptable framework that provides the logic and organization for models and analyses.
Analyzers combine counts, rules, statistics, advanced analytics and machine learning models to solve specific network security and network performance use cases. Because each network is unique, an analyzer can be used either in standard form or customized for specific situations. ThreatEye also provides data caching and a statistical infrastructure to support the development of new analyzers.
In practical terms, analyzers provide an end-user with a programmable interface for running multiple analytics and models in parallel, either on the full stream of network data or on a filtered stream of specific flow types. In addition to running n number of scripts, ThreatEye can ingest both Threat Intelligence data and policy rules as additional input sources for machine learning.
As network dynamics continue to evolve, the flexible nature of ThreatEye’s scripting architecture allows customers to implement new analyzers as new interests and use cases arise.
*ThreatEye refers to CounterFlow AI’s Network Visibility Platform
statistically investigate and classify encrypted network traffic.
- Web Application Identification
- TLS Encryption Protocol Tracking
- OS and Host Fingerprinting
utilize flow information and computed fields to identify anomalous traffic.
- Novelty Detection
- MAC address
- Flow metrics: Protocol, IP
- Domain (TLD, Second Level Domain)
- Layer 7 Mismatch
- PCR DNS
Graph Analytics Analyzers
apply graph theory methods to the incoming stream of connections.
- Graph Clustering
- Communities of Interest ID
enforce specific rules and network policies on incoming network traffic.
- Allowed Servers (DNS, DHCP, NTP, et al)
- Time Fence
- Geo Fence
- TLS Enforcement
Statistical Flow Analyzers
apply statistical models to available flow information and computed fields to identify anomalous traffic.
- Time Anomaly
- Country Anomaly
- DGA Detection
Advanced Anomaly Detection
- Streaming Point Process models
- Naive Bayesian Joint Modeling
(Multi-variable histogram estimation)
look at patterns of traffic over time to identify unexpected bursts and changes in mean and variance.
- Burst/Outlier Detection
- Median Absolute Deviation
- Changepoint detection