Neural networks and game theory improve security in the Internet of Things

Researchers at Baoji University of Arts and Sciences, in China, have developed a new method to defend devices in an IOT environment from wireless network attacks. 

Many devices are nowadays connected to the internet and are capable of collecting, sending and receiving data. This interconnection between devices, referred to as the Internet of Things (IoT), poses some serious security threats, as cyberattackers can target not only computers and smartphones, but also a vast array of other devices, such as tablets, smart watches, smart home systems, transportation systems and so on.

Examples of large-scale IoT implementations could soon become widespread, posing significant risks for businesses and public services that heavily rely on the internet in their daily operations. To mitigate these risks, researchers have been trying to develop security measures to protect devices connected to the internet from wireless network attacks.

The new method combines a deep neural network with a model based on game theory, a branch of mathematics that proposes strategies for dealing with situations that entail competition between different parties.

“Firstly, according to the topology information of the network, the reachability relationship and the vulnerability information of the network, the method generates the state attack and defence map of the network,” the researchers explained in their paper. “Based on the state attack and defence map, based on the non-cooperative non-zero-sum game model, an optimal attack and defence decision algorithm is proposed.”

The method identifies all possible attack and defence paths, and then calculates the probability of success of each of these “attack paths,” a hazard index and the utility value of different attack a defence strategies applicable when the network reaches particular security states. In addition, the interaction between attack and defence is abstracted into a non-cooperative, non-zero and hybrid game model, a game theory framework applicable to problems related offense and defence.

This optimal attack and defence model also integrates prevention and control measures of vulnerable points. The method’s fuzzy system then quantifies an information security risk factor index and feeds it to a radial basis function (RBF) neural network. To optimize and train the parameters of the RBF neural network, the researchers used a particle swarm optimization algorithm. Ultimately, all these steps allow their method to attain an optimized defence model.

In the future, this new technique could help protect IoT devices against wireless network attacks. In a series of simulations evaluating its effectiveness, the defence algorithm performed remarkably well, with an average error below 2 percent.



Read more Innovation Stories