Samrat Ashok Technological Institute

Rapid urbanization and growing vehicle density have led to severe traffic congestion, causing delays, economic losses, and environmental degradation, while traditional traffic prediction models struggle to accurately capture dynamic spatial and temporal patterns. This study proposes a high-performance hybrid traffic prediction framework that integrates a Transformer-based spatial learning mechanism with a Temporal Convolutional Network (TCN) for temporal sequence modeling, optimized using the Adaptive Rectified Adam (AdaRAdam) optimizer to enhance convergence, reduce overfitting, and improve generalization. The model is trained and evaluated on the METR-LA dataset, a widely used real-world traffic speed and flow dataset collected from loop detectors across Los Angeles highways, ensuring realistic benchmarking. Advanced feature extraction, correlation-based feature selection, and ensemble output averaging further enhance predictive stability. Experimental results demonstrate that the hybrid Transformer–TCN model optimized with AdaRAdam achieves 98.7

Abstract Medical imaging has been an essential part of the prior detection and classification of tumors formed in the various parts of the human body. It is known that medical images are of low resolution which hinders the classification or identification of tumors. Usually, tumors will vary widely in their characteristics, behavior, initial position of tumors and prognosis, requiring accurate and timely identification to initiate optimal patient treatment. Till date, breast cancer continues to be the most common cancer diagnosed in women globally, and brain tumors pose significant prognostic difficulties because of their complicated nature and quick progression. A significant amount of research focuses on classifying tumors of specific organs only, but less research contributions in framing a generic framework that can identify and classify different tumors. Hence, in this paper, a generic customized multi-modal approach is devised to identify and classify Breast and Tumors in a single framework, which comprises ResNet-50, VGG-16 and EfficientNet-B0 Deep Learning (DL) architectures. The experimentation work is performed on the MIAS and Br35H datasets for the classification of breast and brain tumors, respectively. To mitigate overfitting and validate robustness, early stopping was employed. Class imbalance in the datasets was addressed through data augmentation. A uniform Hyperparameter setting is applied to all three Deep Learning architectures. EfficientNet-B0 demonstrated superior performance when trained using the Stochastic Gradient Descent (SGD) optimizer along with a learning rate of 0.01, achieving 98.61% accuracy on the MIAS dataset and 98.33% accuracy on the Br35H dataset. While the study focuses on two modalities (digital mammograms and MRI), its adaptability to other tumor types remains a scope for future work.

Segmentation of medical images serves as the indispensable cornerstone of modern diagnostic and therapeutic frameworks, providing the precision required to delineate complex anatomical structures and pathological entities across modalities such as MRI, CT, and ultrasound. The advent of Deep Learning (DL) has fundamentally shifted this landscape, with Convolutional Neural Networks (CNNs) and Transformer architectures achieving unprecedented accuracy. However, the transition from algorithmic innovation to clinical translation is hindered by the “Medical Data Triad” of scarcity, heterogeneity, and opacity. This review presents a comprehensive taxonomic synthesis of the state-of-the-art in “Hybrid Intelligence”—defined here as the synergistic integration of localized inductive biases from CNNs with the global relational modelling of Transformers—to overcome these persistent barriers. Unlike prior surveys, this work provides a critical meta-analysis of “Data-Efficient” methods, including Self-Supervised Learning (SSL), Federated DL, and synthetic data generation via Diffusion models, specifically evaluating their efficacy in mitigating data scarcity while preserving patient privacy. We perform a rigorous cross-comparative assessment of benchmark performance (e.g. BraTS, Synapse, ACDC) against operational constraints such as computational complexity and real-time clinical feasibility. Finally, we bridge the “Technical-Clinical Divide” by mapping regulatory pathways (FDA/EMA) and identifying implementation science gaps, offering a strategic roadmap for the next generation of AI-enhanced, interpretable medical imaging systems.

The paper provides a rather comprehensive overview of movie recommendation systems and demonstrates the importance of these systems in enhancing content delivery in digital entertainment. The models they consider are collaborative filtering, content-based filtering and hybrid architectures. Collaborative filtering is terrific at identifying latent patterns of user behavior, but it falters when the data is sparse, and when there is the cold-start problem. Content-based filtering provides highly personalized suggestions by exploring movie characters, but tends to reduce diversity. As a compromise between personalization, novelty, and robustness, hybrid models attempt to find a happy medium between the two. The paper also explores the key issues surrounding construction of efficient recommendation systems, such as the absence of user interactions, data biases, scaling headaches and instant privacy concerns. They systematically analyze the contemporary streaming services and present the primary performance metrics that measure the effectiveness of a system. The experiments support that optimally designed recommendation systems are significant drivers of enhancing the user experience, engagement, and retention. Finally, the paper recommends the continuous adjustment of the algorithm to enhance flexibility, accuracy, and reliability in dynamic and real-life environments.

Low-data learning is one of the problems of primary concern in computer vision that is generally characterized by large labeled datasets of deep neural networks. The present paper has offered a synthetic data augmented learning framework aimed at enhancing the performance of classification in cases with low sample training. It is an improvement of the baseline convolutional neural networks with the artificial sample generation to improve learning of features. The proposed framework is compared to the known deep learning frameworks such as ResNet18 and AlexNet. Experimentally, synthetic augmentation enhances the baseline performance and deeper residual networks have the highest accuracy. The paper emphasizes the significance of the depth of architecture and diversity of data in low-data settings while maintaining computational feasibility in cloud based environments.