I'm a dedicated Artificial Intelligence enthusiast with hands-on research and internship experience. Proficient in data science, machine learning and big data, with extensive experience in designing and implementing advanced learning algorithms utilising complex neural network architectures. Eager to apply concepts and skills to combat real-world challenges while continuously learning and adapting to technological advancements.
Seeking opportunities.
| S. No | Title | Description | Status | Link |
|---|---|---|---|---|
| 1. | Data Driven Cricket: A Machine Learning Approach to IPL score Prognostication | Implemented and compared the effectiveness of Random Forest, SVM and Linear Regression in predicting T20 cricket scores. | ||
| Random Forest was most effective | Published at the 15th International Conference on Computing and Networking Technology (ICCCNT) held at IIT Mandi · Nov 7, 2024 | IEEE Xplore |
GitHub | | 2. | Perceptually-Aware Image Enhancement with Deep Residual U-Net and PatchGANs | Developed a novel algorithm that can enhance images for versatile cases with minimal training required, with promising evaluation metrics (SSIM: 0.9270, FSIM: 0.9998). | To be Published | GitHub | | 3. | Advanced Photometric Analysis for Predicting Specific Star Formation Rates in Large Galaxies | Explored machine learning (ML) and deep learning (DL) models using SDSS-DR7 photometric data to predict specific Star Formation Rates (sSFRs) in galaxies. Applied models including SVR and Random Forest, achieving a lowest MAE of 0.2413 on original data. Enhanced dataset with synthetic data from Gaussian Copula models for improved robustness. Used DBSCAN clustering to identify high sSFR variability regions. Created an interactive tool for predictions, offering a cost-effective alternative to spectroscopic methods in galaxy evolution studies. | Under review in a journal | Arxive
GitHub | | 4. | Predicting Stellar Metallicity: A Comparative Analysis of Regression Models for Solar Twin Stars | This study employs various regression models, including Random Forest and Linear Regression, to predict the metallicity ([Fe/H]) of solar twin stars using stellar parameters and chemical abundances from the GALAH survey. The Random Forest model excelled, achieving an MSE of 0.001628 and R² of 0.9266, demonstrating the effectiveness of ensemble methods in complex astronomical data analysis. | To be Published | Arxive
GitHub | | 5. | FL-DABE-BC: A Privacy-Enhanced, Decentralized Authentication, and Secure Communication for Federated Learning Framework with Decentralized Attribute-Based Encryption and Blockchain for IoT Scenarios
About this work | This research proposes a secure Federated Learning (FL) framework for IoT, integrating Decentralized Attribute-Based Encryption, Homomorphic Encryption, Secure Multi-Party Computation, and Blockchain. It ensures data privacy and security while enabling decentralized model training, secure aggregation, and real-time analytics in IoT environments. | Published at FMSys: Proceedings of the 2nd International Workshop on Foundation Models for Cyber-Physical Systems & Internet of Things · May 6, 2025 | Read the paper
GitHub | | 6. | Bridging Legacy and Modern Threat Detection: An Analysis of Machine Learning Models on EMBER2018 and CIC-EvasivePDF2022 | Implemented and compared 11 Machine Learning, Deep Learning and Ensemble algorithms for malware detection and classification. Random Forest and XGBoost were the most successful for both legacy (EMBER 2018) malware and modern (CIC EvasivePDF2022) evasive malware achieving well over 99% accuracy. | To be Published | GitHub | | 7. | Multi-Speaker Speech Processing in Noisy Environments: A Hybrid Model for Source Separation and Summarization | Implemented audio source separation using a combination of SepFomers and ConvTas-Net, along with Adaptive noise reduction. Implemented summarization using BART model. | To be Published | GitHub | | 8. | Comparing Analysis of Blackhole Mass Estimation in Type-2 AGNs:Classical vs Quantum Machine Learning and Deep Learning Approaches | This study compares classical and quantum machine learning algorithms for black hole mass estimation in Type-2 AGNs, revealing the strengths and limitations of each approach. Classical models like LSTM excel with high accuracy and error minimization, while quantum models, particularly Estimator-QNN, show promise but struggle with generalization. The results highlight the potential of quantum models for future astrophysical applications as quantum computing advances. | Under Review | Arxive
GitHub | | 9. | Explainable Galaxy Interaction Prediction with Novel Hybrid Attention Mechanisms | This study compares GAN, CGAN, and ContraGAN models for generating synthetic data to predict the B-V color index in astronomy. ContraGAN achieved the best performance with the lowest MSE (0.1294) and highest R² (0.7258), as well as superior clustering. CGANs improved accuracy through feature conditioning, while standard GANs underperformed. The results highlight the value of contrastive learning and conditional generation for scientific data applications. | Ongoing Project | GitHub | | 10. | Adversarial Networks for B-V Index Prediction | This study compares GAN, CGAN, and ContraGAN models for generating synthetic data to predict the B-V color index in astronomy. ContraGAN achieved the best performance with the lowest MSE (0.1294) and highest R² (0.7258), along with strong clustering results. CGANs improved accuracy through conditional generation, while standard GANs underperformed. The findings emphasize the benefits of contrastive learning and conditional models for scientific data synthesis. | Ongoing Project | GitHub | | 11. | Enhancing Image Restoration with Quantum- Integrated Contrastive Adversarial Networks | This study introduces a hybrid image enhancement model combining Contrastive Adversarial Networks (CANs) with Quantum Neural Networks (QNNs) using CUDA-Q for quantum-enhanced feature extraction. The model significantly improves image restoration quality, achieving an average SSIM of 0.8873 and PSNR of 22.5981 on CIFAR-10 images. Quantum integration boosts both detail reconstruction and computational efficiency. Results show strong statistical improvements over noisy baselines (p < 0.0001). | Under review | |