Sir Padampat Singhania University

Most deep learning models face difficulties in analysing image information due to the concept of information bottlenecks and their corresponding methodologies. But, the information bottleneck is used for discarding redundant data and trying to maximise in favour of data directly relevant to the task-oriented information. However, managing information bottlenecks is challenging in the learning model process. Although convolutional neural networks are designed for small-scale processing, their inductive bias makes it difficult to learn contextual features. Thus, we have considered the theoretical learning model to justify the advantages of information bottleneck in deep learning model. We tried to use a fundamental information bottleneck in the vision transformer model. The channel density module cleans up task-related data, while the collected image representations are encouraged to be diverse through local connections in cumulative local transformer blocks. We considered the encoder and decoder methods that analyse the information bottleneck techniques in the deep learning model. This paper presents a rigorous learning theory that mathematically links information bottlenecks for generalisation errors, demonstrating the usefulness of information bottlenecks in deep learning. Our approach suggests that limiting information bottlenecks is crucial for managing errors in deep learning techniques. We conducted experiments across various mathematical models and learning environments to test the validity of our new mathematical insights. In many cases, generalisation errors correspond to unwanted information at hidden levels. We have considered boundary approaches using various scaling parameters and dimensions for the degree of information bottleneck. As per the estimation loss and error by different correlation approaches using generalisation gap methods, we found Spearman correlation having loss (0.86) and error (0.758), whereas Pearson correlation having loss (0.85) and error (0.76), respectively. We also considered outputs for model compression metrics and analysed them through comparative performance.

Breast cancer remains a significant global health concern, with early detection being crucial for effective treatment and improved survival rates. This study introduces HERA-Net (Hybrid Extraction and Recognition Architecture), an advanced hybrid model designed to enhance the diagnostic accuracy of breast cancer detection by leveraging both thermographic and ultrasound imaging modalities. The HERA-Net model integrates powerful deep learning architectures, including VGG19, U-Net, GRU (Gated Recurrent Units), and ResNet-50, to capture multi-dimensional features that support robust image segmentation, feature extraction, and temporal analysis. For thermographic imaging, a comprehensive dataset of 3534 infrared (IR) images from the DMR (Database for Mastology Research) was utilized, with images captured by the high-resolution FLIR SC-620 camera. This dataset was partitioned with 70% of images allocated to training, 15% to validation, and 15% to testing, ensuring a balanced approach for model development and evaluation. To prepare the images, preprocessing steps included resizing, Contrast-Limited Adaptive Histogram Equalization (CLAHE) for enhanced contrast, bilateral filtering for noise reduction, and Non-Local Means (NLMS) filtering to refine structural details. Statistical metrics such as mean, variance, standard deviation, entropy, kurtosis, and skewness were extracted to provide a detailed analysis of thermal distribution across samples. Similarly, the ultrasound dataset was processed to extract detailed anatomical features relevant to breast cancer diagnosis. Preprocessing involved grayscale conversion, bilateral filtering, and Multipurpose Beta Optimized Bihistogram Equalization (MBOBHE) for contrast enhancement, followed by segmentation using Geodesic Active Contours. The ultrasound and thermographic datasets were subsequently fed into HERA-Net, where VGG19 and U-Net were applied for feature extraction and segmentation, GRU for temporal pattern recognition, and ResNet-50 for classification. The performance assessment of HERA-Net on both imaging modalities demonstrated a high degree of diagnostic accuracy, with the proposed model achieving an overall accuracy of 99.86% in breast cancer detection, surpassing other models such as VGG16 (99.80%) and Inception V3 (99.64%). In terms of sensitivity, HERA-Net reached a flawless 100%, indicating its ability to correctly identify all positive cases, while maintaining a specificity of 99.81%, significantly reducing the likelihood of false positives. The model’s robustness was further illustrated through cross-entropy loss convergence and ROC (Receiver Operating Characteristic) curves, with the combined ROC curve showing consistent discrimination ability across training, validation, and testing phases. Overall, the HERA-Net model’s integration of thermographic and ultrasound imaging, combined with advanced deep learning techniques, showcases a powerful approach to breast cancer detection, achieving unprecedented accuracy and sensitivity.

Cashless economies, digital wallets, and electronic transactions have grown in popularity in urban areas due to the influence of various factors. Therefore, the factors that influence the acceptance of e‑wallets among Indian rural consumers remain in question. To bridge this literature gap, the hypothesized model has been developed that integrates the variables of the Technology Acceptance Model, i.e. Perceived ease of use, and Perceived Usefulness, with factors like Digital Financial Literacy, and Perceived Financial Risk to examine their effect on the attitude and intention to adopt e‑wallet services in rural areas of India. The online survey was conducted by using a 7‑point Likert scale using a snowball sampling technique in the 18 administrative divisions headquarters of Uttar Pradesh, India. The final 720 usable responses were evaluated using Structural Equation Modelling (SEM) in SPSS and AMOS v26. The findings indicate that DFL positively influences PU (.514) and PEOU (.689). However, PFR negatively influences PU (‑.372) and PEOU (‑.102). Further, the effect size of PEOU (.740) is greater than PU (.643) on the attitude of consumers. Lastly, attitude significantly influences the intention of adopting e‑wallet services in rural India with a beta value of .617. The study offers several useful suggestions for theoretical and practical implications.

Introduction Conventional Proportional–integral–derivative (PID) controls for multiloop pilot plants are constrained by wired connections, outdated control techniques, and inefficient real-time data sensing and acquisition. As a result, inefficiencies arise, where control loops for parameters like temperature, level, and flow require continuous dynamic adjustments and precise regulation. To overcome this issue, a full wireless solution is proposed which is the need of today’s era. Method This study presents a novel PID controller for a multi-loop pilot plant, utilizing a Bluetooth Low Energy (BLE) based Internet of Thing (IoT) system for wireless, real-time data sensing and control. The system gathers data through sensors and sends it to cloud storage via a BLE access point, which is then monitored using a mobile app called BIP. In addition to monitoring, the BIP app serves as a control interface, allowing PID parameters to be adjusted through a Quantum Firefly- Particle Swarm Optimization (QFPSO) algorithm integrated with the ThingSpeak cloud. This enables the control module to function in three distinct modes for the plant’s loops. Users can manually configure PID parameters, as well as temperature and level set-points, while the system automatically regulates the flow set-point based on real-time data. The BLE-based IoT system comprises five modules using Arduino Nano 33 BLE: a Flow Sensor, a Temperature Sensor, a Level Sensor, IoT communication, and an access point. These modules provide more accurate data than traditional sensing systems. Result Key benefits of the proposed system include wireless accessibility, user-friendliness, a simplified design, ease of upgrades, and consistent control across multiple loops. The proposed system can be easily adapted for various types of industrial control systems with minimal effort. Conclusion Additionally, the developed wireless sensor node can replace wired sensor nodes in any electronic system.

This study investigates numerical pattern formation process in multi-dimensional Gray–Scott reaction and diffusion systems. While many previous studies have been confined to one-dimensional solutions due to the complexities of high-dimensional numerical solutions, this research emphasizes higher dimensions, where pattern formation becomes significantly interesting. Using finite difference and compact difference methods, the study explores pattern dynamics in two- and three-dimensional spaces. It also provides a mathematical foundation and computational analysis of the model from a biological perspective, offering guidelines on parameter selection through linear stability analysis of both non-diffusive and diffusive systems. Numerical experiments showcased remarkable Turing and spatiotemporal patterns.

This paper addresses the pressing issue of diabetes, which is a widespread condition affecting a huge population worldwide. As cells become less responsive to insulin or fail to produce it adequately, blood sugar levels rise. This has the potential to cause severe health complications including kidney disease, vision impairment and heart conditions. Early diagnosis is paramount in mitigating the risk and severity of diabetes-related complications. To tackle this, we proposed a robust framework for diabetes prediction using Synthetic Minority Over-sampling Technique (SMOTE) with ensemble machine learning techniques. Our approach incorporates strategies such as imputation of missing values, outlier rejection, feature selection using correlation analysis and class distribution balancing using SMOTE. The extensive experimentation shows that the proposed combination of AdaBoost and XGBoost shows exceptional performance, with an impressive AUC of 0.968+/-0.015. This outperforms not only alternative methodologies presented in our study but also surpasses current state-of-the-art results. We anticipate that our model will significantly improve diabetes prediction, offering a promising avenue for improved healthcare outcomes in diabetes management.

A modern framework for assessing patient histories and conducting clinical research has been developed as the number of clinical narratives evolves. To discover the knowledge from such clinical narratives, clinical entity recognition and relation extraction tasks were performed subsequently in existing approaches, which resulted in error propagation. Therefore, a novel end-to-end clinical knowledge discovery strategy has been proposed in this paper. The clinical XLNet was used as a base model for handling the discrepancy issue. To predict the dependent clinical relation association, the multinomial Naïve Bayes probability function has been incorporated. In order to improve the performance of the proposed strategy, it takes into account entity pairs presented consecutively through the multi-head attention layer. Tests have been conducted using the N2C2 corpus, and the proposed methodology achieves a greater than 20% improvement in accuracy over existing neural network-based and transformer-based methods.

Heart disease, a leading global cause of death over the past several decades, encompasses a range of disorders affecting the heart. Researchers use various data mining and machine learning techniques to analyze complex medical data, aiding healthcare professionals in predicting cardiac conditions. Despite these advances, existing models often struggle with effectively modelling non-linear relationships, maximizing feature correlation, and addressing challenges related to dimensionality and overfitting. This research paper introduces the Hybrid CCRF model for heart disease prediction, which integrates Canonical Correlation Analysis (CCA) with Random Forest. The proposed model generates polynomial features to capture non-linear relationships and applies Canonical Correlation Analysis to identify canonical variables that maximize correlations between heart disease features and chronic condition features. By combining these canonical variables into a single feature set, the model enhances prediction accuracy. The objectives of the Hybrid CCRF model are threefold: 1) To capture complex non-linear relationships between heart disease and chronic condition features by integrating polynomial feature generation with Canonical Correlation Analysis, thereby improving the model’s ability to represent intricate data patterns; 2) To use CCA to identify and integrate canonical variables that enhance feature correlation, creating a more informative feature set; and 3) To address high-dimensional data and overfitting issues by combining canonical variables with polynomial features in a Random Forest model, balancing complexity and performance for improved generalization and robustness across various datasets. The proposed model achieved an accuracy of 99.45%, with a sensitivity of 98.53%, specificity of 99.54%, precision of 95.73%, and an F1 Score of 0.9711, outperforming all existing models.

Background and objective: One of the most prevalent and significant causes of cancer-related mortality worldwide is considered to be liver cancer. Techniques for automatically segmenting and classifying liver tumors are crucial for supporting medical professionals during the tumor diagnostic process. Classifying liver tumors is challenging due to noise, nonhomogeneity, and the significant appearance diversity observed in tumor tissue. Also, in recent years, most of the research has performed binary classification, but there is still a lack of research on multi-class liver cancer classification. Therefore, we perform a multi-class liver cancer classification and segmentation in this research. Methods: We propose a hybrid deep learning-based multi-class liver cancer classification and segmentation system in this research. The collected CT images are pre-processed in four stages: contrast enhancement, noise filtering, smoothing and sharpening, and liver region segmentation. Next, the binary, texture, histogram, and rotational, scalability, and translational (RST) features are extracted from the pre-processed images. Then, the average correlation (AC) and probability of error (POE) approaches are applied to selecting relevant features and excluding less significant features. After feature selection, a modified AlexNet (MAlexNet) model is used to classify the multi-class classification of liver tumors. Finally, the identified liver tumor regions are segmented using the enhanced U-Net (EUNet) model. Results: Identified liver tumors are accurately classified into three different categories using the proposed classification model: hemangioma (HEM), hepatocellular carcinoma (HCC), and metastatic carcinoma (MET), with an average accuracy of 99.19%. We have collected the multi-class liver cancer CT images from real time patients. The results of the experiments show that proposed hybrid system provides adequate overall accuracy, is less noise-sensitive, and outperforms other state-of-the-art techniques such as SVM, Faster RCNN, Mask RCNN, and SAR-U-Net on a wide range of CT images. Conclusion: Radiologists and doctors can identify liver tumors more accurately using the suggested innovative framework.