Joint Embedding Predictive Architecture (JEPA) world model trained on agent trajectories to predict future latent states from pixel inputs and actions. Uses VICReg loss with RNN dynamics to evaluat…

This project integrates Out-of-Distribution (OOD) detection into self-supervised learning (SSL) to improve rare object representation in long-tailed datasets. It uses a memory buffer to prioritize …

This project is a role-based airline ticketing system supporting customers, agents, and staff, with functionality for flight search, ticket purchases, and flight management. It uses SQL queries and…

End-to-end NBA analytics pipeline for predicting game outcomes and attendance using PyTorch, MLflow, and ONNX. Includes data scraping, model training, quantization, and scalable deployment with Fas…

This project investigates the robustness of parameter allocation strategies in Mix-and-Match (MAM) Adapters for PEFT across different tunable budgets. Our ablation study reveals that optimal alloca…

This project enhances fairness in recommender systems by introducing DNCF, a debiased neural collaborative filtering model capable of handling both binary and multi-subgroup fairness. It builds on …