Computing projects
Implementation and data projects alongside my research.
Accelerated Image Processing and Optimization
Designed and implemented proximal gradient descent, stochastic subgradient methods, ADMM, and coordinate descent for L1/L2 regression, logistic regression, image denoising, deblurring, and matrix completion.
The project emphasized solver performance, accuracy, and convergence behavior. Using Gurobi and careful parameter tuning, experiments scaled to substantially more problem instances while reducing runtime for large-scale problems by up to 60% without sacrificing reconstruction quality.
Model Compression using Adaptive Knowledge Distillation
Developed CIST (Consistently Informative Soft-label Temperature), a knowledge-distillation approach that helps smaller neural networks mimic larger models for faster and cheaper deployment on resource-constrained platforms. See the paper: Consistently informative soft-label temperature for knowledge distillation (arXiv:2605.20357).
CIST achieved up to a 3.63-point accuracy gain over standard knowledge distillation at negligible additional training cost, with results across CNN architectures including ResNet, VGG, WideResNet, MobileNet, and ShuffleNet on CIFAR-100 and ImageNet, and on language-model distillation.
Premier League Ranking Prediction and Disease Symptom Diagnosis
Conducted exploratory data analysis and statistical hypothesis testing, including t-tests and ANOVA, to identify predictive features in structured and unstructured datasets.
Built regression, multi-label classification, and ensemble models including XGBoost, SVM, Naive Bayes, KNN, and K-means from scratch in Python with object-oriented design. Hyperparameter tuning through grid search and cross-validation improved generalization and outperformed Scikit-learn defaults by 3-5% in accuracy.