Federated learning A privacy-preserving ML technique that I was mainly focusing on during my PhD study. Vision-language model, Mixture of Experts, PEFT, Optimization are involved in some works. Curriculum learning An ML technique that mimics human learning process. Knowledge from domain experts guides the learning process in some works. Time series forecasting & parameter estimation Time series forecasting and parameter posterior estimation for wound healing mathematical model. RLHF (PPO) for less toxic summarization Implemented a RLHF (PPO) pipeline for LoRA fine-tuning an LLM to generate less toxic dialogue summarization. LLM fine-tuning Full fine-tuning GPT 4o-mini and Parameter-efficient fine-tuning Llama 3.1 8B. Agentic AI -- deal finder A production calibre Agentic AI solution application for finding best deals with advanced UI. Model Pruning Implementation of three model pruning strategies: magnitude-based individual weights pruning; L1 norm-based filter pruning; and lottery ticket hypothesis pruning. ML at scale on AWS Predicting song popularity using Apache Spark on AWS with S3, EC2 and EMR. Video anomaly detection Privacy preserving fall detection by extracting the pose estimation from videos. Action recognition from pose estimation. Pix2pix dental CT deblurring Deblurring ultra-low dose radiation dental CT for standard dose image quality. Super Resolution for Flow Dynamics Super resolution CNN for synthesized flow dynamics data.
