This review provides a comprehensive overview of the state-of-the-art methods of graph-based networks from a deep learning perspective. Graph networks provide a generalized form to exploit non-euclidean space data. A graph can be visualized as an aggregation of nodes and edges without having any order. Data-driven architecture tends to follow a fixed neural network trying to find the pattern in feature space. These strategies have successfully been applied to many applications for euclidean space data. Since graph data in a non-euclidean space does not follow any kind of order, these solutions can be applied to exploit the node relationships. Graph Neural Networks (GNNs) solve this problem by exploiting the relationships among graph data. Recent developments in computational hardware and optimization allow graph networks possible to learn the complex graph relationships. Graph networks are therefore being actively used to solve many problems including protein interface, classification, and learning representations of fingerprints. To encapsulate the importance of graph models, in this paper, we formulate a systematic categorization of GNN models according to their applications from theory to real-life problems and provide a direction of the future scope for the applications of graph models as well as highlight the limitations of existing graph networks.

This paper proposes a hand free human-machine interaction (HMI) system to establish a novel way for communication between humans and computers. A regular interaction system based on the computer mouse puts the user’s hand for too long in a pronation posture that increases inflammation in the wrist and hand. Additionally, the need for hand obstructs the use of computers for handicap people. In this paper, we develop a new pointing device for differently able people based on open and closed human eyes with inertia measurement that restrict to deal with carpal tunnel syndrome (CTS) for regular people and enables a novel way to interact with computers for the handicap people. The proposed system carries the human head gesture and eyes to perform the movement and clicking event of the mouse cursor. A combined three-axis accelerometer and gyroscope is used to detect the head gesture and translate it into the position of the mouse cursor on the computer monitor. To perform the left and right-clicking event, the user needs to shut down the left and right eye for a moment while opening another eye. This paper is also carried out the design of a deep learning approach to classify the individual openness and closeness of both human eyes with quite a high accuracy of 95.36% that ensures the comprehensive control over the clicking performance. The use of complementary filter removes the noise and drift from the obtained performance and confirms the smooth and accurate operation of the proposed device. An experimental validation is added to show the effectiveness of the proposed HMI system. The experimental details along with the performance evaluation prove that the proposed HMI system has extensive control over its performance for differently able people.

Faults in a power electronic component (inverter, converter, modules etc.) can severely affect the reliability, efficiency, as well as security of the entire power conversion system if not detected by early warning. The inverters: being the most common power converters in the industries are widely used in power supplies, energy management systems, motor control, etc. Nevertheless, due to the diverse operating conditions and complexity of the drive system, the power converters employed in drive systems are more prone to faults. Moreover, collecting a large amount of labeled data in the non-linear industrial environment is not practical and this imposes challenges to the prior approaches as they claim a bulk amount of labeled data to be generalized. Taking this as a research challenge, this paper proposes a sparse autoencoder based unsupervised power converter fault detection and classification scheme which learns the fault-oriented features from the unlabeled dataset. In addition, the precision and sensitivity study are conducted in this paper to justify whether the proposed scheme delivers a reliable performance or not.

This paper depicts a remote vision-based manual motion control of five degrees of freedom (DoF) articulated industrial robotic arm intended for teleoperation. Remotely operated robot arms are now involved in handling heavy loads, detrimental materials or to explore uninhabitable spaces. In this light, a robotic system is developed in this paper to administrate this manipulation tasks from a distance in order to improve job safety and to alleviate the function of professionals. The proposed system carries a modular mechanical design for simplifying the function and maximizing the performance of the robot arm. The kinematic design of the 5 DOF robotic arm is carried out first to develop a sophisticated model that reflects the motion in the real world. The verification of the kinematic model is performed by observing the output response obtained from the simulation study. The robotic system considered in this study is a fixed base robotic system where the position of the robotic manipulator is determined by the operator. Before conveying the developed system to the real world, the computer-aided design is accomplished. Thereafter, the Live camera feeds are taken by a wide-angle camera, and the data bits are transmitted to a base station. A human operator then manipulates the robot arm using a controller according to the received camera feeds. An IEEE 802.11 standard communication technique is adopted to establish a three-way communication path. The paper presents the mechanical design along with the kinematic analysis, simulation studies and its application to real world explicates the comprehensive control over its performance.

This paper presents a novel knowledge distillation neural architecture leveraging efficient transformer networks for effective image classification. Natural images display intricate arrangements encompassing numerous extraneous elements. Vision transformers utilize localized patches to compute attention. However, exclusive dependence on patch segmentation proves inadequate in sufficiently encompassing the comprehensive nature of the image. To address this issue, we have proposed an inner-outer transformer-based architecture, which gives attention to the global and local aspects of the image. Moreover, The training of transformer models poses significant challenges due to their demanding resource, time, and data requirements. To tackle this, we integrate knowledge distillation into the architecture, enabling efficient learning. Leveraging insights from a larger teacher model, our approach enhances learning efficiency and effectiveness. Significantly, the transformer-in-transformer network acquires lightweight characteristics by means of distillation conducted within the feature extraction layer. Our featured network's robustness is established through substantial experimentation on the MNIST, CIFAR10, and CIFAR100 datasets, demonstrating commendable top-1 and top-5 accuracy. The conducted ablative analysis comprehensively validates the effectiveness of the chosen parameters and settings, showcasing their superiority against contemporary methodologies. Remarkably, the proposed Transformer-in-Transformer Network (TITN) model achieves impressive performance milestones across various datasets: securing the highest top-1 accuracy of 74.71% and a top-5 accuracy of 92.28% for the CIFAR100 dataset, attaining an unparalleled top-1 accuracy of 92.03% and top-5 accuracy of 99.80% for the CIFAR-10 dataset, and registering an exceptional top-1 accuracy of 99.56% for the MNIST dataset.