Profile
Education
Experience
Publications
Languages
Collapse all
Expand all
Light mode
Profile
NLP Research Resident at VinAI Research, advised by Assoc. Prof. Thien Huu Nguyen.
I work on LLMs (inference, truthfulness, trustworthy, safety). I work on Information Extraction.
I work on Continual Learning.
I work on LLMs (inference, truthfulness, trustworthy, safety). I work on Information Extraction.
I work on Continual Learning.
Education
B.S. Computer Science GPA: 4/4 (Rounding to the Nearest Integer 🚀)
Oct 2023
Hanoi University of Science and Technology, Hanoi, Vietnam
+ Notable courses taken:
- Linear Algebra
- Probability and Statistics
- Machine Learning and Data Mining
- Information Retrieval
- Deep Learning and Its Applications
- Data Science
- Fundamentals of Optimization
- Programming Techniques
- Computer Architecture
- Evolutionary Computation
- Object Oriented Programming
- Data Structures and Algorithms
- Discrete Mathematics
Experience
Research Resident
VinAI Research, Hanoi, Vietnam
Feb 2023 - Present
- Conducted research in Large Language Models
- Conducted research in Continual Information Extraction
- Participated in weekly meetings of the NLP reading group
- Attended courses relevant to research interests
Publications
Asterisk (*) denotes equal contribution.
We introduce a realistic dataset for safety evaluation of LLMs which is more competent in exposing toxicity in LLMs than state-of-the-art datasets.
ToVo: Toxicity Taxonomy via Voting
Tinh Son Luong*, Thanh-Thien Le*, Thang Viet Doan*, Linh Ngo Van, Thien Huu Nguyen, Diep Nguyen
NAACL 2025 | Model
We introduce a method for effortless, automatic creation of customized moderation tools, and an accompanying dataset named ToVo.
We propose a novel multi-task learning framework to effectively handle catastrophic forgetting in Continual Relation Extraction.
SharpSeq: Empowering Continual Event Detection through Sharpness-Aware Sequential-task Learning
Thanh-Thien Le*, Viet Dao*, Linh Van Nguyen*, Thi-Nhung Nguyen, Linh Ngo Van, Thien Huu Nguyen
NAACL 2024 | Code
We design a Sharpness-Aware framework to tackle the the Sequential-Task challenge of Continual Event Detection.
PhoWhisper: Automatic Speech Recognition for Vietnamese
Thanh-Thien Le, Linh The Nguyen, Dat Quoc Nguyen
Tiny Paper @ ICLR 2024 | Model
We fine-tune the Whisper model on an 844-hour dataset that encompasses diverse Vietnamese accents.
Lifelong Event Detection via Optimal Transport
Viet Dao*, Van-Cuong Pham*, Quyen Tran*, Thanh-Thien Le, Linh Ngo Van, Thien Huu Nguyen
EMNLP 2024 | Preprint | Code
We propose aligning the classifier module to the pre-trained language modeling head via optimal transport to enhance Continual Event Detection.
We design a retrieval-based paradigm, leveraging LLM-generated descriptions to address Few-shot Continual Relation Extraction, boosting performance and reducing forgetting.
We improve prompt-based Continual Relation Extraction by incorporating task-specific prompt pool that capture the variations within each task while maximizing cross-task variances.
Languages
- Programming Languages: Python, C/C++
- Libraries/Modules: PyTorch, NumPy/SciPy, pandas, matplotlib, seaborn, cvxpy
- Language Proficiency: English (fluent), Vietnamese (native)