Islam Hegazy

Recent advancements in generative artificial intelligence have boosted significant interest in conversational agents. The visual dialog task, a synthesis of visual question-answering and dialog systems, requires agents capable of both seeing and chatting in natural language interactions. These agents must effectively understand cross-modal contextual information and generate coherent, human-like responses to a sequence of questions about a given visual scene. Despite progress, previous approaches often required complex architectures and substantial resources. This paper introduces a generative dialog agent that effectively addresses these challenges while maintaining a relatively simple architecture, dataset, and resource requirements. The proposed model employs an encoder-decoder architecture, incorporating ViLBERT for cross-modal information grounding and GPT-2 for autoregressive answer generation. This is the first visual dialog agent solely reliant on an autoregressive decoder for text generation. Evaluated on the VisDial dataset, the model achieves promising results, with scores of 64.05, 62.67, 70.17, and 15.37 on normalized discounted cumulative gain (NDCG), rank@5, rank@10, and the mean, respectively. These outcomes underscore the effectiveness of this approach, particularly considering its efficiency in terms of dataset size, architecture complexity, and generation process. The code and dataset are available at https://github.com/GhadaElshamy/MS-GPT-visdial.git , complete with usage instructions to facilitate replication of these experiments.

In the agricultural sector, a huge threat is posed by harmful insects and plant diseases. Hence, it is important to conduct early diagnosis and detection of such diseases. Detecting plant diseases is now possible with the great aid from continuously developing deep learning methods which represent a robust tool, rendering remarkably meticulous results. However, the deep learning models’ accuracy is dependent on the labeled training data quality and volume. Accordingly, this paper proposes a deep learning-based method for detecting tomate disease, combining recurrent neural network (RNN) architecture with the convolutional neural network (CNN) for this purpose. Xception-GRU is the proposed model as it begins with the Xception pre-trained model and is followed by the GRU layers. After that, transfer learning is employed for training the Xception-GRU model on real and synthetic images for tomato leaves images to be classified into 10 disease categories.Three different classifiers are used on the features extracted from the Xception-GRU model. These classifiers are multi-layer perceptron (MLP), support vector machine (SVM), and the k-nearest neighbor (KNN). With extensive testing and training on PlantVillage dataset, available publicly, the proposed model reached 100%, 98.79%, 99.85%, and 100% accuracy in classifying tomato leaf image diseases into (Early, Late Blight and Healthy), (Early and Late Blight), (Healthy and Late Blight) and (Early and Healthy). Hence, it is shown that the approach proposed is superior over the present methodologies.

Smart agriculture holds a transformative potential, driven by cutting-edge technologies, in revolutionizing the global food production sector and enhancing food safety measures. This is achieved by leveraging the capabilities of smart maps, artificial intelligence, and agricultural data analytics. Hence, farmers can benefit from it allowing them to streamline resource utilization. This, in turn, leads to reduced inputs of fertilizers, labor, seeds, and water, while simultaneously amplifying crops. Therefore, this paper introduces a comprehensive system for advancing smart map agriculture, comprising five pivotal stages designed to yield accurate and meaningful results. Notable progress is achieved through meticulous data acquisition by employing the rice seedling and Weed Net datasets, coupled with feature extraction utilizing the Mobile Net architecture. The classification stage is critical for distinguishing between rice and land in the first dataset and between crops and weeds in the second. It further enhances the system’s efficacy. Moreover, segmentation using K-means clustering to detect rice or crop is employed. Notably, the current approach exhibits promising performance levels. It results in 99.7%, 100%, 99.4%, and 99.72% for accuracy, precision, recall, and F1_score, respectively, in the rice seedling dataset, as well as achieving 98.72%, 97.3%, 100%, and 98.63% in the WeedNet dataset. The primary contribution of this study lies in the integration of MobileNet and SVM architectures for feature extraction and classification in the agriculture field. It also presents the innovative use of K-means clustering for segmenting agricultural fields.

Internet of Things (IoT) networks are integral to modern life due to their pervasive connectivity and automation capabilities. Intrusion Detection Systems (IDS) are crucial in IoT ecosystems to countermeasure attacks that can compromise devices and disrupt essential services. Their role is vital in maintaining the integrity, confidentiality, and availability of data within these networks. The effectiveness of these security systems is fundamentally dependent on the robustness of learning algorithms and the quality of the datasets utilized. Class imbalance is a common challenge in real-world datasets, where certain classes are represented by significantly fewer instances compared to others. This paper studies the impact of balancing the BoT-IoT dataset on the performance of Machine Learning (ML) based IDSs using three algorithms: K-Nearest Neighbors (KNN), Gradient Boosting (GB), and Support Vector Machine (SVM). To address the class imbalance problem, we apply two resampling techniques, random upsampling and Synthetic Minority Over-sampling Technique (SMOTE). We evaluate the efficacy of the models through various performance metrics, including accuracy, precision, recall, and F1-score. The findings of our experimental work prove that balanced datasets lead to more dependable and robust IDSs that are capable of handling real-world data with varied class distributions.

The proliferation of Internet of Things (IoT) devices has become inevitable in contemporary life, significantly affecting myriad applications. Nevertheless, the pervasive use of heterogeneous IoT gadgets introduces vulnerabilities to malicious cyber-attacks, resulting in data breaches that jeopardize the network’s integrity and resilience. This study proposes an Intrusion Detection System (IDS) for IoT environments that leverages Transfer Learning (TL) and the Convolutional Block Attention Module (CBAM). We extensively evaluate four prominent pre-trained models, each integrated with an independent CBAM at the uppermost layer. Our methodology is validated using the BoT-IoT dataset, which undergoes preprocessing to rectify the imbalanced data distribution, eliminate redundancy, and reduce dimensionality. Subsequently, the tabular dataset is transformed into RGB images to enhance the interpretation of complex patterns. Our evaluation results demonstrate that integrating TL models with the CBAM significantly improves classification accuracy and reduces false-positive rates. Additionally, to further enhance the system performance, we employ an Ensemble Learning (EL) technique to aggregate predictions from the two best-performing models. The final findings prove that our TL-CBAM-EL model achieves superior performance, attaining an accuracy of 99.93% as well as high recall, precision, and F1-score. Henceforth, the proposed IDS is a robust and efficient solution for securing IoT networks.

Recently the conjunction between vision and language has created many intersecting tasks as visual question-answering systems, image captioning, etc. Specifically, dialog systems that depend on a visual scene play an important role in improving human-computer interaction technology. At the same time, reinforcement learning has emerged as a very successful paradigm for a variety of machine learning tasks, especially those tasks that aim to develop smart and humanoid machines. In this paper, we show how reinforcement learning is applied to conversational agents to build a powerful visual dialog agent. Visual Dialog task requires the agent to have a meaningful conversation about visual content in natural language. For a given image, its caption, dialog history (question/answer pairs), and a question about this scene, the agent should comprehend the question, extract the relevant context from the history, and ground this information on the image to correctly answer the current question. Two main visual dialog tasks have been introduced which are a free-form dialog task known as “Visual Dialog” and a goal-oriented dialog task formulated as a guessing game. Two datasets have been introduced to address these tasks which are VisDial dataset and GuessWhat?! datasets. For evaluation, some approaches use the accuracy metric while others use four metrics that have been proposed for the sake of this task. Several approaches are proposed for tackling this task based on supervised learning or reinforcement learning or even combining both techniques. This paper represents a comparative study of eleven important reinforcement learning approaches for visual dialog.

Cancer is one of the diseases, caused by cell divisions, that can be fatal. It has been the second most common cause of death for the past period of years globally. Any region of the body can be affected by a wide range of disorders collectively referred to as cancer. Neoplasms as well as malignant tumors are other terms that are used for describing cancer. Skin is one of the body parts that can be affected. Early melanoma skin cancer detection is a must so that the mortality rate of skin cancer patients is decreased. The accuracy of early detection of melanoma skin cancer can be enhanced through applying machine learning methods. This paper provides a model that can detect melanoma skin cancer early. This model is used using the dataset that the International Skin Imaging Collaboration has provided. The proposed model using support vector machines achieved a promising accuracy of 95.96%.

A longer battery life is a highly sought-after feature for most smartphone users when considering their next device. However, with the emergence of new hardware technology and software applications that require heavy processing, the demand for battery power has significantly increased. Unfortunately, the development of battery technology has not kept up with the rapid advancements in smartphone hardware and software, which rely heavily on battery power. To address this issue, several approaches have been proposed to regulate battery consumption and the charging process on smartphones. In this paper, we summarize the different approaches related to this problem that managed to achieve up to a 61% increase in battery daily usage in simulation testing, highlighting their strengths, limitations, and current challenges. Furthermore, we provide a comprehensive review of various open-source datasets that have the potential to be used in developing new approaches to improve battery drainage and degradation in smartphones. We also discuss the methodology for collecting each dataset. Finally, we propose a new approach to address the current limitations and challenges to solving the problem of battery drainage and degradation that could be developed using the currently available datasets. These new approaches may involve incorporating machine learning techniques to predict battery charging patterns and minimize battery drainage

With the exponential rise of Internet of Things (IoT) devices in our modern-day lifestyle, the potential danger of botnet intrusions has become inevitable. These botnet attacks cause extensive damage to both individual users and large organizations. In the realm of IoT botnet intrusion detection, learning-based Intrusion Detection Systems (IDS) were proposed as a powerful strategy for identifying malicious behaviors and consolidating the security of IoT systems. Moreover, Transfer Learning (TL) techniques have emerged to overcome the problems of scarce data and the long training time of traditional learning approaches. This paper proposes an IDS for IoT networks based on the TL EfficientNet model. We train the system and validate it using the Bot-IoT network intrusion dataset. Our approach incorporates balancing the dataset, redundancy removal, dimensionality reduction, and transformation of the dataset into RGB images to be fed as input to the detection model. The findings of our experimental work prove that the proposed model is robust and effective, and it surpasses existing models with an accuracy of 99.53%, recall of 99.53%, and F1-score of 99.52%.