Research Goal: Collect continuous data with lab test ground truth to create an ML model diagnosingTuberculosis infection that if robust across situationally diverse conditions.◦ Data Processing: Using techniques to filter out ambient noises and normalize signals across the different sample.

◦ LLMs: Studied different LLMs’ transformer architectures, such as LLaMA2, LLaMA3, CodeLlama, Falcon,Mistral, Mixtral, Breeze, Whisper, and Vicuna.◦ LoRA: Researched Low-Rank Adaptation (LoRA) methods for efficient fine-tuning of large models.◦ MoE: Investigated Mixture of Experts (MoE) techniques to optimize transformer model performance.◦ RAG: Explored Retrieval-Augmented Generation approaches for improved contextual information retrieval.◦ ZeRO: Worked with aiDAPTIV2 (Phison AI SSD) in both Model and Data Parallel processing.◦ Parameters Calculation Project: Calculate the size of different LLMs models’ weights and activations by the config file (link: https://pingw220-llm-memory-calculation-app-w9m7pa.streamlit.app/)

◦ Processed MIDI Data: Processed instruments’ sounds into MIDI data and converted them into a training set formachine learning.◦ AI Music Interaction: Researched real-time interaction with AI music, led by graduate and PhD students.◦ Arduino MIDI Receiver: Created Arduino MIDI receiver to both collect data and test the model.

Published Research: Published to Journal of Student Research (Feb. 2023), DOI: 10.47611/jsrhs.v12i1.4037.◦ Electronic Structures: Investigated changes in electronic structures and sizes of hexagonal boron nitride (hBN)particles through various spectroscopy techniques.