Benchmarking Fairness Metrics for Optimizers Designed robust experiments to evaluate the fairness and performance of different optimization algorithms (SGD, AdamW, and Muon). Tested models across a variety of tasks, ranging from linear regression to imbalanced image classification and Large Language Models (SmallGPT2)
Movie Genre Classification Benchmarked the performance differences between a standard CNN (ResNet18) and a modern Vision-Language Model (Qwen3). Optimized both approaches to run efficiently under strict, low-budget hardware constraints using a single L4 GPU.