Trang . Email and Phone Number

My main research is the tiny models, efficient models that can deploy on edge devices. Recently work is the Neural Architecture Search (NAS) for High-performance Image Classification (Computer Vision).Achievement:- Journal Publish: "Zero-Shot Proxy with Incorporated-Score for Light-Weight Deep Neural Architecture Search," coauthored with T.-T. Nguyen and J.-H. Han is accepted in Electronics, 2024- Conference paper: "ZenNet-SA: An Efficient Lightweight Neural Network with Shuffle Attention for Facial Expression Recognition" is accepted in ICCAS, 2024. (10/2024)

Main responsibilities1, Joint Intent Detection and Slot Filling for Vietnamese: In this project, we research on IDSF for Vietnamese. We reached a new SOTA on the PhoATIS dataset at that time. Our models' performance reached 97.89 % for Intent accuracy and 88.68 % for Sent Acc. Our work was published at RIVF 2022

Main responsibilities: 1) Optimizing notification volume based on ML model: In this project, I collect and analyze users' data. I researched and proposed a model that helps give an efficient budget notification to users based on Linear optimization. I provided API and then tested on a real news system, and the number of notifications was dramatically reduced by 30 % without reducing user engagement. 2) Recommendation favorite post on the news system based on ML model: In this project, my tasks consist of processing and analyzing user data from distributed systems and then applying the ML model to real data. This model can suggest the favorite post to users based on the user's history data. Moreover, my modules can update and process user's data and re-train at scheduled times automatically. I also provided API and then tested it on a real news system. 3) Text Summarization for Vietnamese on News data: In this project, I conduct research on Text summarization for Vietnamese. Interestingly, we demonstrate that only a simple filtering technique is required to generate training data of sentence-summary pairs without any human annotation. Our approach can generate text with 92% Syntax correctness equal to over 80 % Factual correctness (evaluation by human). Our work was published at PACLIC 2020