Internet of Things (IoT) makes it possible to utilize a multitude of small- and medium-scale devices allowing for increased functionality in flexible networks. However, user authentication on IoT devices is as important as it can be challenging. Due to limitations on the available energy, interface, and processing power, IoT devices can be the weakest link in their networks. Multiple authentication techniques have been developed to address these challenges. However, the existing techniques are limited in terms of performance, overheads, and efficiency. In contrast, our proposed authentication method uses a user’s walking gait as an input, because gait is unique to every user and can be collected using low-power inertial sensors found on all handheld devices. Our authentication method uses a lightweight neural network (NN) which is further complimented with early exits to further optimize computational cost(s). We also propose reinforcement learning that considers the energy consumption to dynamically determine which of the exits should be chosen to strike a balance between performance and computational and energy cost. Though effective, one of the challenges with IoT devices is their power supply such as dependence on batteries. Discharge of the battery or any other interruptions can lead to recomputations, which are expensive on already limited battery-operated IoT devices. To address such challenges, especially for user authorization, we introduce intermittent computation to our proposed authentication framework. Intermittent computation can store the state of the NNs at checkpoints. In the case of power disruption, the execution will be resumed from the saved checkpoints instead of performing the whole execution. Most implementations of intermittent computation take place at the compiler level, which makes for a very efficient design, however, that also makes them hardware specific. Our method is implemented on software and is hardware-agnostic, allowing us to create checkpoints to save and retrieve the authentication framework state, in case of power interruption. The proposed authentication framework can authenticate users with up to more than 85% accuracy and can save up to 34% of computations due to the proposed intermittent computing.

The increase in the deployed Internet-of-Things (IoT) devices has facilitated better functionality and connectivity across devices. Authentication of users on IoT devices plays a key role in the IoT networks to ensure security and integrity of the data. Multiple user authentication techniques including cryptographic and biometric approaches are introduced for authentication of users on these devices. Despite their effectiveness, these techniques incur large computational and communication overheads. In contrast, we propose a gait-based authentication, suitable for IoT devices in this work. Across multiple gait signals, we consider walking gait in this work, as it is unique to every individual and can be measured in an unobtrusive manner by utilizing the inertial sensors, which are inherently embedded in the IoT devices as well as smartphones. Given the limited resources available on IoT devices, we propose a lightweight authentication method that allows for early exit from the Neural Network (NN) in order to optimize the computational costs. A Deep Q-Learning Network (DQN) reinforcement learning method is further introduced to determine the exit dynamically during the authentication. The proposed method has been evaluated on the whuGAIT dataset. The proposed technique achieves more than 85% authentication accuracy with $6.94\times$ lower inference time and $5.9\times$ reduction in multiply-and-accumulate operations compared to ResNet50.