As more and more devices rely on data sharing, the Internet of Things (IoT) is rapidly expanding. Developing strong privacy and security protocols for these systems is critical, as devices, applications, and communication networks become increasingly integrated. However, as these systems grow, cyber attacks on the IoT are becoming increasingly complex. These attacks often involve machine learning, earning themselves the name “smart attacks”.
The security solutions necessary to combat these attacks are computationally heavy and frequently involve a large communication load. Additionally, due to their relatively small computational abilities, many IoT devices are more vulnerable to attacks than computer systems. For example, outdoor IoT devices, such as the sensors used for low-grade security protection, are frequently designed to be low-cost. Due to their computational limitations, however, these devices often have difficulty detecting a cyber attack early enough to stop it.
Despite the practical challenges of implementing effective security protocols in IoT devices, the protection of user privacy and data remains critical. Machine learning-based approaches have been used to implement important security systems, including IoT authentication, access control, secure offloading, and malware detection methods. In fact, these machine-learning approaches often help to overcome the computational limitations of IoT devices, enhancing their security while maintaining or decreasing their onboard computational requirements.
One vital security component, authentication, allows IoT devices to distinguish between source nodes and outside attacks. In order to maintain low consumption of computational power, authentication techniques typically provide security protection by focusing on the features of radio channels and transmitters in the physical (PHY) layer of the device. These PHY-layer characteristics are then compared to the characteristics identified by the transmitter. To determine whether a transmission is authentic, the characteristics are compared to a threshold, intended to maximize the accuracy of the classification. Because the environment of the IoT is unpredictable, however, it can be difficult for an IoT device to choose an appropriate threshold to maximize this accuracy.
Machine learning techniques, such as Q-learning, can be used to select an optimal threshold to achieve the highest accuracy of authentication. Q-learning is a reinforcement technique that finds an optimal authentication policy that results in the best possible outcome. When selecting the optimal threshold, Q-learning identifies a threshold that maximizes the number of attack transmissions correctly identified as attacks, while minimizing the number of authentic transmissions incorrectly identified as attacks.
Other machine learning techniques can also be incorporated into IoT authentication. Supervised learning techniques, like Franke-Wolfe or incremental aggregated gradient can be used to improve spoofing resistance, while unsupervised learning techniques such as infinite Gaussian mixture model can be used to authenticate nearby devices while still securing information related to the location of the device. Deep learning techniques can be used to improve the authentication accuracy still farther, although these techniques are more computationally intensive and require greater resources.
With access control, IoT devices prevent the access of resources by unauthorized users. This process is made more difficult in IoT devices because networks are frequently made up of various types of data from a variety of sources. Machine learning techniques like support vector machines, k-nearest neighbors, and neural networks have been used to detect unauthorized users. Due to the complexity of this type of security, computational limitations often constrain the security of lower-grade IoT devices.
Access control protocols can be developed using machine learning techniques and then executed in the IoT devices as a simple decision process. This technique of managing access control reduces the energy and computational power required by the system, and, as a result, helps to increase the device’s lifespan. For example, a k-nearest neighbor method can be used to identify outliers among the data, providing a method to identify any unauthorized users. Additionally, a multiplayer perceptron can calculate a “suspicion factor’ denoting the likelihood that an IoT device may be a target of attacks. Other machine learning techniques, such as support vector machines, have been used to detect attacks on Internet traffic and electricity grids.
Secure IoT offloading
With secure offloading, IoT devices are able to use external, cloud-based computation and storage resources for tasks that require heavy computational power or for which latency must be minimized. Similarly to how Q-learning can be used to identify the optimal threshold for IoT authentication, Q-learning can be used to identify the optimal rate of offloading data to combat jamming and spoofing attacks.
In this scenario, the Q-function determines the long-term reward of the offloading data rate based on the power jamming the system, the importance of the task, the channel bandwidth, and the channel gain. Other machine learning techniques can further improve performance. For example, given enough computational power, a deep Q-network technique can accelerate the learning speed past that of Q-learning alone.
IoT devices must be able to detect malware to prevent threats to security and performance. Supervised learning techniques can be used in IoT devices to detect malware by identifying atypical behavior. For example, one malware detection method uses k-nearest neighbors to cluster network traffic and then uses random forests to identify malware among the regular traffic. In order to limit the load on the IoT device and to achieve faster computational speed, this malware detection can also be offloaded to a server.
To maximize malware detection performance, Q-learning can select the optimal threshold for offloading to maximize accuracy while minimizing latency and energy consumption. Another supervised learning technique, the Dyna-Q method, can also be used to find the optimal offload threshold. This method has been shown to increase accuracy and decrease latency when compared to the Q-learning approach.
Securing IoT devices has become increasingly difficult as larger quantities of data are being shared through the network using relatively low-level technology. Cyber attacks are becoming increasingly complex and are more difficult to combat, especially when using hardware with limited computational ability. However, the protection of user data remains critical in the face of these threats.
Despite the challenges of implementing security protocols in IoT devices, machine-learning techniques provide a method of implementing important security systems such as IoT authentication, access control, secure offloading, and malware detection schemes. Machine learning techniques may be the key to securing the IoT.
Daniel Browning is the Business Development Coordinator at DO Supply Inc. In his spare time, he writes about automation, AI, technology, and the IoT