Publications
* denotes equal contribution. · Up-to-date list on Google Scholar.
2025
-
Pre-printCAMELLIA: Benchmarking Cultural Biases in LLMs for Asian LanguagesGeorgia Tech · Samsung R&D · Sungkyunkwan University · Tohoku University · NUS · University of Copenhagen · TakeNote.ai · University of Michigan · IIT Delhi · Institute of Science TokyoAs Large Language Models (LLMs) gain stronger multilingual capabilities, their ability to handle culturally diverse entities becomes crucial. Prior work has shown that LLMs often favor Western-associated entities in Arabic, raising concerns about cultural fairness. In this paper, we introduce CAMELLIA, a benchmark for measuring entity-centric cultural biases in nine Asian languages spanning six distinct Asian cultures. CAMELLIA includes 19,530 entities manually annotated for association with specific Asian or Western culture, as well as 2,173 naturally occurring masked contexts for entities derived from social media posts. Using CAMELLIA, we evaluate cultural biases in four recent multilingual LLM families across tasks such as cultural context adaptation, sentiment association, and entity extractive QA.
2024
-
Book Chapter · IGI GlobalCharting the Ethical Course: Navigating AI Advancements in Communication EducationIn Sanae Elmoudden & Jason S. Wrench (Eds.), The Role of Generative AI in the Communication Classroom
-
Book Chapter · IGI GlobalGenerative AI: Ethical Challenges for Teachers and Learners — Striking a BalanceIn Shalin Hai-Jew (Ed.), Generative AI in Teaching and Learning
2023
-
Book Chapter · IGI GlobalChatbots as Motivational Agents: Chatbots — The Value of a Digital Tool in PedagogyIn Mohammad Amin et al. (Eds.), Trends, Applications, and Challenges of Chatbot Technology
2021
-
Conference · ICICT 2021MAGNeto: An Efficient Deep Learning Method for the Extractive Tags Summarization ProblemThe 7th International Congress on Information and Communication Technology (ICICT), 2021. PIXTA Vietnam & Hanoi University of Science and Technology.In this work, we study a new image annotation task named Extractive Tags Summarization (ETS). The goal is to extract important tags from the context lying in an image and its corresponding tags. We adjust state-of-the-art deep learning models to utilize both visual and textual information. Our proposed solution combines convolutional and self-attention layers with a novel gating mechanism, auxiliary loss functions, and an unsupervised pre-training strategy. Our model achieves 90% F1 score on the public NUS-WIDE benchmark and 50% F1 score on a noisy large-scale real-world private dataset.
2018
-
Symposium · Panasonic 2018An Enhanced X-Vector Model for Noise-Robust, Text-Independent Speaker IdentificationPanasonic Technology Symposium, 2018. Panasonic R&D Center Vietnam.We develop an in-house, low-cost, scalable solution for closed-set text-independent speaker identification. We propose three directions to enhance accuracy and noise-robustness: (1) retraining on large open-source noisy datasets, (2) omitting excessive data augmentation, and (3) a new algorithm to detect non-voice samples. Our enhanced model achieves a 9% increase in identification accuracy over existing state-of-the-art open-source solutions and achieves 97% non-voice detection accuracy.
2014
-
Workshop · ComNavi-14Monitoring Scintillation Effects over Vietnam by Means of a GNSS Software ReceiverWorkshop on Communications and Navigation for the Development of Vietnam's Marine Economy, 2014.
2013
-
Conference · IEEE ICL-GNSS 2013Recent Results in Receiving and Decoding Signals from the Beidou System2013 International Conference on Localization and GNSS (ICL-GNSS). IEEE.Since December 27, 2012, the Beidou Navigation Satellite System officially started to operate. This event is a great opportunity for researchers in South East Asia to receive and analyze Beidou signals. After the official statement, researchers at NAVIS Centre monitored the broadcasted signal using NAVISOFT — our Software Radio Receiver. This paper shows the analysis of the navigation message broadcasted by Beidou satellites on the B1I bandwidth. We were able to observe valid ephemeris data on visible satellites and demonstrate successful PVT computation using combinations of GEO and MEO/IGSO satellites in static conditions.