Project Overview
✅ Certification on CITI Social & Behavioral Research -Human Subjects
✅ Physionet credentialing for Dataset Access
✅ Dataset Access MIMIC-III
RAG Language Model : Performance Study
Skill-Based Model-Based Reinforcement Learning (Skimo) on Maze
Implemented SKIMO (Skill-Based Model-Based Reinforcement)in a Maze environment, achieving the best loss of 14 after training on 3046 trajectories, enhancing complex task efficiency.
Developed a joint training strategy for skill dynamics using 256-length samples over 750 iterations, significantly improving sample efficiency, Validated SKIMO on long-horizon tasks, with consistent objective achievement and a steady 1–point loss drop every
100 iterations, demonstrating model effectiveness.
Prompt Engineering on a Dialogue Summarization Task
Fine-tuned the Flan-T5 model, achieving absolute percentage improvements of the instruct model over the original model:ROUGE1: 0.26%, ROUGE-L: 0.26%, ROUGE-Lsum: 0.26%.• Conducted prompt engineering using the Flan-T5 model and DialogSum dataset, exploring the impact of different prompts and comparing zero-shot and few-shot inferences, Defined training arguments and created a Seq2SeqTrainer instance for the model, utilizing a higher learning rate than full fine-tuning for enhanced performance.
Performing Prompt engineering on a dialogue summarization task using Flan-T5 and the dialogsum dataset. Exploring how different prompts affect the output of the model, and comparing zero-shot and few-shot inferences.
Project Overview :
