Software Defined Networking (SDN) is aimed at rethinking network topologies and mitigating the limits that have emerged in traditional networks. It is a relatively new networking architecture that has become the most widely discussed networking technology in recent years and the latest development in the field of developing digital networks, which aims to break down the traditional connection between the control level and the infrastructure level surface. It allows you to fix a number of legacy mesh issues as well as add a whole host of new functionality to your existing networks. The goal of this separation is to make resources more manageable, secure, and controllable. As a result, many controllers such as Beacon, Floodlight, RYU, OpenDayLight (ODL), Open Network Operating System (ONOS), NOX, & POX have been developed. Selection of the finest-fit controller has altered to an application-contingent operation due to the large range of SDN applications and controllers. The data circulate allying the application level and the data level surface is managed by a network controller using Southbound Application Program Interfaces (APIs) and Northbound APIs. Before proceeding down to the switches and making logical judgements, the counteract determines the majority of the forwarding decisions. Advantages include worldwide control and observation of the entire network at once, useful for automating operations such as network operation, better server and network utilization, and so on. This paper discusses SDN, a new networking model in which the architectonics transitions from a completely distribution to a greater extent centralized form, and evaluates and contrasts the effects of various SDN controllers on SDN. The fundamental component of the control plane that oversees all data plane operations is the controller. As a result, its capabilities and performance (SDN controllers such as POX, Ryu, ONOS, ODL, etc.) are crucial in maintaining optimal performance. There are many controller suggestions available in the paper, and this study compares them to see which is the best overall it will be used by the SDN community as a reference for various SDN controllers. This paper presents performance analysis of various controllers and work done by different researchers in the last 9 years.

Text classification is an emerging area in Natural Language Processing (NLP). On the other hand, traditional text classification methods need to be improved due to the complexity and semantic nature in text. In this paper, we build a hybrid deep learning model using Deep Learning (DL) and Machine learning (ML) models. This work combines two traditional neural networks namely Long Short-Term Memory (LSTM) and Recurrent Neural Network (RNN) to extract the features from the text document. LSTM preserves the historical information for text sequences and extracts the features using the RNN structure. The extracted features are used to run on machine learning classification algorithms like AdaBoost and XGBoost to perform the final prediction. Thereby the proposed Deep Hybrid Model eliminates the fully connected classification layers from a typical Deep Learning model. The performance of proposed model is measured with other models and the results show that the deep hybrid model provides about 12% increased results in terms of accuracy in text classification.

Instagram is one of the leading social media networks with millions of users accessing this platform on a daily basis. From a business point of view, it is a billion-dollar industry, especially with paid promotional posts. The visibility of an Instagram post is dependent on the user’s social ties to the platform and the hashtags associated with the post. In this paper, we have proposed a category/genre-based hashtag recommendation system for Instagram posts. The proposed system uses pre-trained EfficientNet trained on “imagenet” weights for deep feature extraction and an Applied Least Squares (ALS) model for generating hashtags. Transfer learning is used with the base model of pre-trained EfficientNetB2 to enhance the feature extraction process. We further compare the performance of our model with similar recommendation models by replacing the EfficeintNetB2 with other popular CNNs (Resnet-50, Resnet-169, and Inceptionnet-V4). Our proposed model with EfficeintNetB2 achieves better precision, recall, and F1-measure values compared to other similar models using CNNs as feature extractors.

The centrifugal motor pump unit, our case study, is a rotating machine, which thanks to a bladed rotor, increases the kinetic energy and projects, using centrifugal force, the liquid at the periphery on the volute. At the exit and with the help of a divergent, a large part of the kinetic energy is transformed into driving pressure. This equipment currently occupies a prominent place in water pumping stations. These installations are considered as a main ring in the chain of hydraulic circuits. They have an important role in the extraction, transformation, and transfer of water between different points. However, without effective maintenance, this equipment cannot render an acceptable service and the populations will not derive the expected benefits. In addition, the equipment will have a very limited lifespan and the investments will be wasted. As a result, an effective maintenance plan is essential for the proper functioning and proper operation of the pumping installations. The aim of this study is to assess the availability of critical elements of a motor pump system after 24 months in service at a water pumping station located in the city of Constantine in eastern Algeria. In this study, we'll use the graphical tool SADT (Structured Analysis Design Technique) to identify the various subsystems and components of the motor pump system, a Pareto chart analysis and then a failure mode, effect, and criticality analysis (FMECA). This analysis helps us to investigate failure patterns associated with each component in order to choose the appropriate plan of action.

The intelligent factory defined by Industry 4.0 is established on intelligent machines, services, and production. To cope up with these requirements, topics such as Artificial Intelligence (AI), Evolutionary Computation (EC), Internet of Things (IoT), and Big Data are gaining a lot of attention in the manufacturing sector. In spite of the use of optimized input parameters, owing to some unknown moments, a machining activity tends to produce tool wear, break, and chatter that affect tool life, resulting in poorer surface roughness. Thus, there is a need to adopt self-monitoring of tools so that the diagnosis can be done without any human intervention. As the 4th industrial revolution pervades production sector, the volume of manufacturing data created has approached big data proportions, and is very dynamic. In an attempt to compute such vast data, Deep Learning (DL) approach is being considered to be a potential tool. In this paper, investigation of unknown vibration moments leading to cutting tool faults is ventured upon through spectrogram and deep learning ensemble i.e. LSTM (Long-Short-Term Memory)-Auto encoder.

A DDoS attack is similar to a Denial-of-Service attack in that harmful traffic is generated from various sources and orchestrated from a single point. Distributed Denial of Service attack prevention is substantially more difficult than DoS attack prevention from a single IP address since the traffic sources are scattered, typically all over the world. The main aim of this research paper is to provide a novel framework in which DDoS can be detected at an early stage by making use of a machine learning model and then proper mitigation methods can be taken. For experimentation, SNORT, an open-source access tool is used and TCP-SYN DDoS traffic is generated using hping 3 tool towards ONOS SDN controller. Our proposed method detects the DDoS attack in early stages and traffic from the destination end is blocked as soon it detects the malicious traffic.

Cognitive biases are the primary source of errors in decision-making. These biases enable learners to make snap judgments, leading to errors. These biases usually result in incorrect decisions when students solve Multiple-Choice Questions (MCQ) and Short-Answer Questions (SAQ) types of problems. Furthermore, each person's brain is unique with respect to cognitive processing; their perception capability and biases may vary when confronted with the same situation. Hence, it is critical to investigate learners' numerous biases. This article analyzes various types of biases, such as Anchor effect bias, Availability bias, Attention bias, Confirmation bias, and Framing effect bias when students are asked to solve MCQ and SAQ types of problems. Students' performance in an introductory course on programming is used to analyze the effect of cognitive biases on their performance. We recorded audio interviews with respondents to ascertain why they selected a particular option for a given question. Three experts manually classify the respondent's biases in response to a specific question using this audio clip.

The increasing and global spread of Coronavirus (COVID-19) has made facemasks imperative and valuable. It established new norms to our way of life with regulations that are necessary for survival. This study portrays the methodological significance of image processing using Deep Learning: MobileNet-v2 cascade for detection of the masked face and spawning face embedding. It achieves the best results for larger datasets as MobileNet-V2 is a convolutional semantic network with a depth of about 53 layers, meanwhile, the application of similar methods on smaller datasets proves challenging. This paper paves a path of exploring detection on the basis of the Single Shot Detector (SSD) algorithm that introduces a channel attention mechanism to improve the ability of the model to express salient features while simultaneously utilizing information of different feature levels optimizing the function loss. It also sheds light on the resultant output, which creates a large chunk of data categorized as big data. The algorithm shows final experimental results predicting the goal of face recognition and mask detention as successful and effective with an accuracy of the results ranging between 90-95%.