Hi,
I'm Sai Harsha Varma
i am into
About Me
I am a technologist and hardcore programmer with excellent analytical skills, known for my discipline and process-oriented mindset. My passion lies in solving real-time problems and constructing highly scalable software systems using Artificial Intelligence and Machine Learning. I thrive on the challenge of breaking down complex issues, meticulously analyzing each aspect of the problem, and developing innovative solutions. Collaborating with diverse teams, I strive to make a meaningful impact and drive positive change through technology.
email : saiharshavarma@outlook.com
place : New York, USA
Master of Science | Computer Science
Bachelor in Technology | Computer Science and Engineering
11th and 12th | Telangana State Board of Intermediate Education
High School | ICSE
May 2023 - November 2023
May 2023 - August 2023
October 2022 - November 2022
May 2022 - July 2022
December 2021 - Present
October 2021 - November 2021
Abstract: Tropical cyclones are considered to be one the most devastating natural disasters, that pose severe threats if not accounted for before their arrival. Predicting the intensity of such tropical cyclones becomes critically important for effective disaster management and response. However, existing techniques of Dvorak's estimation do not take into account the physical attributes of a cyclone contributing to its intensity and assume a steady state for the same failing to incorporate the temporal aspect to them. This study proposes the use of a Nadam-optimizer based Hierarchical Attention Enhanced Multihead Convolutional Neural Network, with additional pre-processing of level sets segmentation applied. Unlike the aforementioned existing methods, this approach accounts for all physical factors of influence in a cyclone that level sets segmentation leverages to extract contours and cyclonic patterns. Subsequently, the Hierarchical Attention-Enhanced Multihead CNN itself focuses on these highlighted features to extract intensity aided by the Nadam optimizer which adapts to any further noise during the gradient estimates in the prediction. Additionally, for further focus on the Region of Interest, the Indian Ocean as per the used dataset of (IR) satellite images from INSAT3D Imagery and cyclone wind speed details from IBTrACS, coordinate-based cropping was applied for the cyclonic extraction. Therefore, the proposed approach establishes itself as a robust tropical cyclone intensity estimation technique, whilst providing superior performance in comparison with other potential deep learning methods like those of AlexNet, CycloneNet, ResNet, LeNet, and such.
Abstract: Efficient and accurate fire and smoke detection systems are essential to prevent potential catastrophes, save lives, and minimize property damage. This research paper introduces an innovative solution that combines deep learning principles, specifically leveraging the InceptionV3 convolutional neural network architecture, with real-time surveillance for a groundbreaking approach to early fire and smoke detection. The proposed methodology incorporates an attention mechanism to enhance the model's discriminative ability by focusing on salient features within visual data. By harnessing deep learning's capability to extract intricate patterns from extensive datasets, the InceptionV3 model identifies subtle visual cues indicative of fire and smoke instances. Augmented with the attention mechanism, the model effectively prioritizes relevant regions of interest, resulting in elevated detection accuracy. Real-time surveillance further accelerates the system's effectiveness through continuous environmental monitoring, expediting the detection-to-response timeline. In experiments, the proposed architecture achieves remarkable training and validation accuracies of 99.64% and 97.44%, respectively. Comparative assessments against baseline models confirm the superiority of the proposed approach. This research contributes to the advancement of fire safety by integrating cutting-edge technologies, fostering safer environments, and enhancing disaster management strategies.
Abstract: This research explores the essential domain of breast cancer detection by leveraging the capabilities of machine learning algorithms. Breast cancer, a critical global concern, necessitates early identification for effective patient outcomes. This study merges machine learning and medical research for more precise breast cancer detection. Our approach employs machine learning algorithms to differentiate malignant and benign breast cancer cells using quantifiable nuclei attributes from fine needle aspiration images. The dataset contains 569 samples, each with 32 features. Algorithms include AdaBoost, XGBoost and K-Nearest Neighbors with comprehensive data preprocessing including scaling and handling missing values. Techniques like PCA and SMOTE handle class imbalance and high dimensionality. Across scenarios, XGBoost consistently outperforms AdaBoost and KNN in accuracy, precision, recall, and F1-score. This research underscores the potential of machine learning in enhancing breast cancer diagnosis, contributing to advanced methodologies that emphasize the significance of quantitative nuclei attributes in accurate detection.
Abstract: The present research endeavors to introduce a groundbreaking web application, which serves as a unified billing management system specifically designed for pharmacies. The current state of pharmacy management systems is characterized by a fragmented approach, where separate software solutions are utilized for employee management and e-commerce functionalities. This fragmented approach has led to various inefficiencies and a subsequent rise in operational costs. One significant constraint of existing systems is found within the arduous process of managing inventory. The proposed solution aims to mitigate the issue by incorporating an advanced search capability that encompasses various filters. This feature enables the swift identification of medicines, thereby ensuring the precise fulfillment of prescriptions. The present system has been meticulously developed with the primary objective of identifying expired pharmaceutical products. This innovative feature serves to bolster the overall safety of medication by effectively thwarting the inadvertent inclusion of expired medicines in customer orders. Moreover, the web application integrates sophisticated deep-learning techniques for the purpose of inventory management, facilitating the option of either manual data entry or image-based identification. In summary, the present study introduces a transformative solution for the management of pharmacies, effectively tackling the drawbacks associated with fragmented software systems. The application presented in this study aims to provide a secure and efficient platform for pharmacies by integrating billing, inventory, and customer interactions. Additionally, the application utilizes deep learning techniques to facilitate inventory updates. This research paper explores the potential benefits of this unified approach and discusses how it can contribute to a streamlined and enhanced future for pharmaceutical management.
Abstract: Molecular Dynamics (MD) simulations provide qualitative insights into the dynamics of liquids, solids, and liquid-solid interfaces under varying temperature and pressure conditions. Accurate force calculations for ion cores are crucial for the success of classical MD simulations. This research explores the application of MD simulations to study large-scale systems and understand atom structure and interactions. MD simulations can successfully examine protein dynamics, such as folding and unfolding, contributing to a better understanding of their behaviour. Integrating MD simulations with experimental data enables a holistic examination of atomic-level properties and their impact on cellular behavior. Additionally, MD simulations offer valuable insights into protein-ligand interactions and facilitate drug development. This study introduces a innovative optimization technique that uses verlet algorithm with OpenMP to parallelize MD simulations, resulting in significant computing time reductions for force and energy evaluations. This optimization methodology shows promise in various domains, including protein-ligand interactions and complex system investigations. The results show that, when compared to the serialisation simulation, the proposed approach can deal with computational load balancing challenges better and more effectively, while also reducing computing time.
