news

Latest news and announcements about Tran Phuoc Thanh Nhan's research, projects, and achievements.

Sep 18, 2026 Paper published at MAPR 2026: Retrieval-Augmented Fine-Tuning with Reasoning Distillation for Vietnamese Medical Question Answering (with Huynh Nguyen Phu, Truong Quoc Truong, and Dang Van Thin). The paper proposes a retrieval-augmented fine-tuning (RAFT) pipeline for Vietnamese medical QA that distills Chain-of-Thought rationales from a large language model into LLaMA-3-8B via Low-Rank Adaptation (LoRA). Using oracle-and-distractor contexts built from VMHQA and a fixed oracle-context protocol for retrieval-independent evaluation, the fine-tuned model reaches 93.64% accuracy and 93.56% macro-F1, outperforming both the provided-context baseline and zero-shot LLaMA-3-70B while remaining competitive with prior VMHQA results including fine-tuned GPT-4o.
Jul 01, 2026 Paper accepted at SemEval-2026 Task 9: Gradient Descenders at SemEval-2026 Task 9: Data-Centric Counterfactual Augmentation for Multi-Label Hate Speech Detection (with Dang Van Thin). The work addresses multi-label hate speech detection by introducing a data-centric counterfactual augmentation strategy that generates targeted training perturbations. The approach improved robustness and generalization in the shared task evaluation, with findings presented at the 20th International Workshop on Semantic Evaluation (ACL).
Jun 15, 2026 Paper accepted at DeSegMa-IT: Gradient Descenders at DeSegMa-IT: Leveraging Monolingual Transformer for LLM-Generated Text Detection and Boundary Identification (with Bui Hong Son and Dang Van Thin). The paper tackles detection of LLM-generated text by leveraging a monolingual Transformer, jointly addressing classification and boundary identification of machine-written spans. The approach was evaluated within the DeSegMa-IT shared task at the International Workshop on Evaluation of Natural Language and Speech Tools for Italian.