Adrino Rosario James

I am an undergraduate student in computer science at Christ University, specialising in machine learning and artificial intelligence. I maintain a First Class standing with an 82% average in my first year.

My research interests focus on:

• Computer Vision: I aim to develop efficient architectures for visual recognition tasks, exploring the tradeoffs between model capacity, computational efficiency, and generalization. My work spans classical feature engineering to modern deep learning approaches, with emphasis on practical performance and model interpretability.

• Classical Statistical Learning: I am particularly interested in feature extraction, probabilistic reasoning, and model interpretability. I believe that understanding the foundations of statistical learning is crucial for developing robust and explainable AI systems.

• Efficient Deep Learning: I investigate hybrid architectures that combine different neural network building blocks (residual connections, dense connections) to achieve strong performance with reduced parameters. This facilitates deployment in resource-constrained environments while maintaining competitive accuracy.

Relevant Coursework: Data Analysis using R, Discrete Mathematics, Artificial Intelligence and Human Machine Interface, Research Methodology, Data Structures and Algorithms

Research Experience and Projects

Stellar and Galactic Object Classification - Independent Research (2025)

Conducted independent research on astronomical object classification using the Sloan Digital Sky Survey dataset (5M+ observations). I developed machine learning models for multi-class astronomical object classification and engineered domain-specific features from photometric and spectroscopic data, achieving 96% classification accuracy with classical ML algorithms—placing in the top 5% performance benchmark. I conducted systematic comparison with loss functions and hyperparameter tuning to optimize model performance.

Efficient Deep Learning Architectures for CIFAR100 - Independent Research (2025)

Designed and evaluated hybrid CNN architectures combining residual blocks and dense connections for fine-grained image classification (100 classes). I achieved 60% top-1 accuracy while reducing model parameters by 30-50% compared to standard ResNet baselines through architectural innovations. I investigated tradeoffs between model capacity, computational efficiency, and generalization through systematic comparisons.

Transfer Learning vs Training from Scratch on Small Datasets - Independent Research (2025)

Conducted an empirical study comparing transfer learning approaches against custom CNN architectures for image classification with limited data. I evaluated preprocessing strategies against architectures trained from scratch and analysed trade-offs in generalization, training stability, and accuracy while working with small-scale datasets.

Technical Skills

• Programming: Python (Numpy, Pandas, ScikitLearn, Seaborn, SciPy), R, SQL, C++, Java
• ML/DL Frameworks: PyTorch (torchvision)
• Specializations: Computer Vision
• Tools & Platforms: Git, GCP, Jupyter, LaTeX

Research Skills

• Research Design: Hypothesis formulation and testing, sampling techniques, survey design, critical evaluation of research literature
• Academic Writing: Literature review synthesis, abstract writing, scientific paraphrasing, citation management (Zotero)
• Research Ethics: IRB protocols, ethical research practices, data privacy considerations

Teaching and Contributions

Contact: adrinorjames@gmail.com — +91 90084 21698
LinkedIn: linkedin.com/in/adrinorosario
GitHub: github.com/adrinorosario