The world wide use of the public internet has opened up the possibility of applying similar technology for automated functioning of physical objects equipped with micro electronic devices. The perception is to apply that kind of technology to various areas of human activity for enhancing the quality and scale of performance. Starting from automated homes, the possibility of application are projected to be in efficient transport systems, energy management, smart cities, industrial automation, environmental monitoring, business management, defense operations etc. Such activities require huge network support, deployed over vast areas of the globe with suitable taxonomy for fast connectivity and operation. In this paper, it is suggested that the network is hierarchical having hybrid tree graph architecture, with the root node as the overall controlling headquarter. The hierarchy of the nodal stations is assumed to be based on the geographical distance from the root node, in which the nodes having the same hierarchy are also linked by collateral bus connectivity. The search of a node and routing data to the destination is carried out along a path which is as close as possible along the geographical direction of the location of that destination. Such a procedure was earlier developed by this author [19], [20], accounting fully for the sphericity of the globe. Accordingly, an algorithm for the search and routing in the envisaged network is presented in conclusion.

In mobile ad-hoc networks (MANET), nodes are randomly distributed and move freely, and hence the network may face rapid and unexpected topological changes. The Greedy Perimeter Stateless Routing (GPSR) Protocol is a position-based routing protocol in MANET, in which the greedy forwarding strategy is used for data forwarding. In this strategy, the next hop is selected greedily among the one-hop neighbors, that is closest to the destination. This mechanism may cause route breakage since the farthest neighbor is more likely to step out of the transmission range of the forwarding node, and hence the packet loss rate is high. Also, periodic broadcasting of beacon packets to maintain neighbors’ positions leads to higher control traffic overhead in the network, and this consumes network bandwidth, leading to an increase in network congestion.\nIn this paper, an improved greedy perimeter stateless routing protocol, called “LDAB-GPSR”, is proposed. LDAB-GPSR mainly focuses on maximizing the packet delivery ratio while minimizing the control overhead. In order to accomplish this, two techniques are introduced, the first one is the location prediction technique in which the greedy forwarding strategy is improved by choosing more stable routes for data forwarding. The second one is the adaptive beaconing technique in which the slow start algorithm is employed to adapt the beacon packet interval time based on the mobility of nodes and the data traffic load instead of using the periodic beaconing strategy. These two strategies together improve the overall performance of the GPSR routing protocol.\nThe performance of the new proposed protocol is evaluated by carrying out several NS-2.35 simulation experiments. The simulation results show that LDAB-GPSR protocol significantly outperforms both the original GPSR and GPSR-PR protocols. In terms of packet delivery ratio, LDAB-GPSR produced results 40% better than GPSR and 29% better than GPSR-PR. In terms of control traffic overhead, LDAB-GPSR achieved 25% improvement over both the GPSR and GPSR-PR protocols. Throughput of LDAB-GPSR was 40% better when compared with GPSR and 29% better when compared with GPSR-PR. In terms of average end-to-end delay, LDAB-GPSR was 68% better when compared with GPSR and 20% better when compared with GPSR-PR.

Deep Learning is essential in machine learning. Deep Learning has two main models: Convolutional Neural Network is used to feature extraction in image processing, and Recurrent Neural Network is used to handle sequence recognition. Deep Learning is applied in many fields, such as image processing, natural language, automated systems, and virtual assistants. The biggest difficulty when using Deep Learning models is that the input data of the model is very complex and big data. In contrast, machine learning algorithms often use linear transformations and only perform well when data is distributed on the plane. In the case of spatially rotating objects such as the motion of marker or robotic arms, the data of these objects can be distributed in the form of spheres or hyper-spheres, so the linear methods have the results not high. In this paper, we propose using a Conformal Geometric Algebra (CGA) mathematical tool to feature extraction for the Long Short Term Memory (LSTM) model. This paper proposes feature extraction using CGA to reduce dimensions and create the feature vector. Then, the CGA feature vector for the input data of the LSTM model to human activity recognition. The Experiments show that the proposed method using CGA based on LSTM has better classification results than using Principal Components Analysis based on LSTM.

In this paper, a complex non-linear programming problem with the two parts (real and imaginary) is considered. The efficient and proper efficient solutions in terms of optimal solutions of related appropriate scalar optimization problems are characterized. Also, the Kuhn-Tuckers conditions for efficiency and proper efficiency are derived. This paper is divided into two independently parts: The first provides the relationships between the optimal solutions of a complex single objective optimization problem and solutions of two related real programming problems. The second part is concerned with the theory of a multi objective optimization in complex space.

Answering questions about visual content may be a research area to create a man-made intelligence system which may answer questions on a picture. If the system is provided with an image and a natural language question, the task of the system is to provide answers for the given question. Since we are humans and we have common sense, is it easy for us to look at the image and answer any queries regarding that image. However, a visually-impaired person can\'t draw out information from the image. To help the visually impaired, both the questions and answers should be open-ended. An Artificial intelligence system that can solve this task should demonstrate a more generalised understanding of images: it must be able to answer completely different questions about an image, also it should be able to even address different sections of the image. Visual questions selectively target different areas of an image, such as the colour of a book, object in the background.

The recent advancement in the pattern recognition technique has demonstrated the superiority in remote sensing technology, where the deep neural network uses the spatial feature representation such as convolution neural network (CNN), to provide better generalization capability. Regardless of any CNN structure, the prediction always involves uncertainty and imprecision while classifying the ultra-high resolution (UHR) image. Our aim is two-fold: firstly, increase the reliability feature by performing the multiscale fusion via a Markovian network (MN) called MCNN-MN (or first-stage fusion). Secondly, the additional information is processed using a multi-layer perceptron (MLP) framework and integrated with focal-information (MCNN-MN) to rectifies the unreliable information using Constraint-based Dempster Shafer theory(C-DST), also known as the MN-MLP method (second-stage fusion). The proposed MN-MLP framework exploits the complementary information from the classification map using reliability (or confidence) function at the regional level. The robustness of the proposed method was tested on two different labelled datasets. One of the data was acquired using the drone survey in the semi-urban region of Dhanbad district, India, and labelled with building features. Another is the well-known Potsdam dataset with five land cover classes (Tree, building, grassland, impervious surface, and car). The proposed framework (MN-MLP) outperformed the MLP, MCNN-MN, and other baseline methods.

Although many endpoint security mechanisms have been built and stacked up, still numerous attacks are being vectored resulting in maintaining a profile of network attacks. Even though studies do uncover the evil nature of the attacks, there exist various forms that carry those evil senses. Sometimes, network administrators even do not know that an attack has happened from the inside network because of heavy data traffic. If they want to analyze those data packets, administrators must be in place. Existing mechanisms handle the nature of protecting networks by considering the source, destination IP addresses, source, and destination port addresses. On the other hand, a transparent analysis of packets helps in the detection of an attack. In our project, we use YARA for robust rule-based detection to hunt packets that are thrown into the network to cause an attack. To deal with the detection and analyzing of packets on the go, we propose an automated system in which the task of packet capturing and as well as the analysis of those packets using YARA is done. YARA is the name of a tool primarily used in malware research and detection. This level of operation introduces a new endpoint mechanism where the network pcap files are sent through the YARA rules to detect and identify what network traffic may constitute, and the results are alerted to the network administrator through mailing services so that the administrator can take fine action to secure the network.

After the everyday use of systems and applications of artificial intelligence in our world. Consequently, machine learning technologies have become characterized by exceptional capabilities and unique and distinguished performance in many areas. However, these applications and systems are vulnerable to adversaries who can be a reason to confer the wrong classification by introducing distorted samples. Precisely, it has been perceived that adversarial examples designed throughout the training and test phases can include industrious Ruin the performance of the machine learning... This paper provides a comprehensive review of the recent research on the threat of Axis in multiple variant fields. We classify ourselves among the early researchers in that it is also imperative to note that the paper focuses only on the recent documents published during 2018-2021. The diverse systems models have been investigated and discussed regarding the type of attacks, some possible security suggestions for these attacks to highlight the risks of adversarial machine learning.