This post is also available in:
עברית (Hebrew)
Software-defined networks (SDNs) have become a cornerstone of modern data centers, cloud infrastructure, and enterprise environments. By separating the control layer from the traffic-forwarding layer, SDNs make networks easier to manage and reconfigure. However, that same centralized architecture can create vulnerabilities. If attackers overwhelm or compromise network controllers, disruptions can quickly cascade across large portions of the network.
Researchers have now developed an AI-driven security system designed to identify malicious traffic while simultaneously preventing network controllers from becoming overloaded during an attack. The approach combines a deep quantum neural network with an optimization method inspired by the collective behavior of wild geese and dwarf mongooses.
The first component focuses on threat detection. According to TechXplore, incoming network traffic is analysed by a deep quantum neural network, a machine-learning model designed to identify complex patterns and anomalies that may indicate malicious activity. The system specifically targets abnormal traffic flows associated with attacks such as distributed denial-of-service (DDoS) campaigns, where large volumes of requests are used to overwhelm network resources.
Detection alone, however, is only part of the solution. In software-defined networks, controllers act as centralized decision-makers. If one controller becomes overloaded while handling suspicious traffic, network performance can degrade even before the attack succeeds.
To address that problem, the researchers added a second layer based on bio-inspired optimization algorithms. Drawing inspiration from how wild geese coordinate movement and how dwarf mongooses organize group behavior, the system continuously evaluates controller workloads. When a controller begins approaching capacity limits, network switches can be reassigned automatically to other controllers with available resources.
This combination allows the platform not only to identify attacks but also to redistribute workload dynamically, helping maintain network stability while the threat is being handled.
In simulations, the system achieved a detection accuracy of 93.7 percent, with a true positive rate of 91.6 percent and a true negative rate of 87.5 percent.
From a cybersecurity perspective, the approach reflects a growing trend toward integrating threat detection and network resilience into a single framework. Rather than treating security and traffic management as separate functions, future network defenses may increasingly combine AI-based attack detection with automated infrastructure adaptation, helping organizations respond to attacks before service disruptions spread through critical systems.
The research was published here.
