papers

2026

1. PRO (AAAI-26)

   

   Probabilities Are All You Need: A Probability-Only Approach to Uncertainty Estimation in Large Language Models

   Manh Nguyen, Sunil Gupta, and Hung Le

   The 40th Annual AAAI Conference on Artificial Intelligence, 2026

   TL;DR: Estimates predictive entropy from top-K output-token probabilities with an adaptive K; +2.4% avg AUROC over the best baseline across 3 QA datasets.

   arXiv Bib Code Poster

   @inproceedings{nguyen2026probabilities,
     title={Probabilities Are All You Need: A Probability-Only Approach to Uncertainty Estimation in Large Language Models},
     author={Nguyen, Manh and Gupta, Sunil and Le, Hung},
     booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
     year={2026},
     pages={32546--32554}
   }

2. RAD (ECML 2026)

   

   Retrieval-augmented Decoding for Improving Truthfulness in Open-ended Generation

   Manh Nguyen, Sunil Gupta, and Hung Le

   2026

   TL;DR: Builds a tiny reference store of truthful logits from  10 examples to steer decoding; +3.2% avg truthfulness on TruthfulQA, no retraining.

   arXiv Bib Code Poster

   @inproceedings{nguyen2026rad,
   title     = {Retrieval-Augmented Decoding for Improving Truthfulness in Open-Ended Generation},
   author    = {Nguyen, Manh and Gupta, Sunil and Le, Hung},
   booktitle = {Machine Learning and Knowledge Discovery in Databases. Research Track},
   year      = {2027},
   publisher = {Springer Nature Switzerland},
   pages     = {19--36}
   }

3. RCS

   

   Efficient Best-Of-N with Radial Consensus Score

   Manh Nguyen, Sunil Gupta, and Hung Le

   2026

   TL;DR: Ranks sampled LLM answers by distance to their embeddings’ semantic center; +2-7% selection accuracy over strong baselines incl. majority voting.

   arXiv Bib Code

   @article{nguyen2026rcs,
     title     = {Efficient Best-of-{N} with Radial Consensus Score},
     author    = {Nguyen, Manh and Gupta, Sunil and Le, Hung},
     journal   = {Advances in Neural Information Processing Systems - Workshop: Linguistic Principles for Foundation Models (LP4FM)},
     year      = {2026},
     note      = {arXiv:2604.12196}
   }

4. UAB

   

   Uncertainty-Aware Budget Allocation for Adaptive Test-Time Reasoning

   Manh Nguyen, Sunil Gupta, and Hung Le

   2026

   TL;DR: Reallocates a fixed sampling budget toward harder questions using early-round disagreement as a difficulty signal; +2.3% avg accuracy over uniform allocation (up to +5.5% peak).

   arXiv Bib Code

   @article{nguyen2026uncertainty,
     title={Uncertainty-Aware Budget Allocation for Adaptive Test-Time Reasoning},
     author={Nguyen, Manh and Gupta, Sunil and Le, Hung},
     journal={arXiv preprint arXiv:2605.26849},
     year={2026}
   }

5. DAR

   

   Hear Both Sides: Efficient Multi-Agent Debate via Diversity-Aware Message Retention

   Manh Nguyen, Anh Nguyen, Dung Nguyen, and 2 more authors

   2026

   TL;DR: Each round, broadcasts only the most disagreeing agent responses instead of every message, cutting multi-agent debate noise while keeping useful diversity.

   arXiv Bib Code

   @article{nguyen2026hear,
     title={Hear Both Sides: Efficient Multi-Agent Debate via Diversity-Aware Message Retention},
     author={Nguyen, Manh and Nguyen, Anh and Nguyen, Dung and Venkatesh, Svetha and Le, Hung},
     journal={arXiv preprint arXiv:2603.20640},
     year={2026}
   }

6. TB-MAD

   

   Think More or Read More? A Token-Budget Theory of Reasoning with Shared Context

   Manh Nguyen, Sunil Gupta, and Hung Le

   2026

   TL;DR: Predicts the optimal number of debate agents from single-agent measurements alone; self-consistency matches or beats debate in 93% of 980 settings under tight token budgets.

   arXiv Bib

   @article{nguyen2026think,
     title = {Think More or Read More? A Token-Budget Theory of Reasoning with Shared Context},
     author = {Nguyen, Manh and Gupta, Sunil and Le, Hung},
     year = {2026},
   }

7. VCA

   

   Volatility-Clustering Adaptation for Financial Time Series

   Manh Nguyen, Minh Hoang Nguyen, Huu Hiep Nguyen, and 2 more authors

   2026

   TL;DR: Adds an autocorrelation penalty on squared returns to the training loss so rollouts match real volatility clustering; improves over the pretrained model, strongest under FORE evaluation.

   arXiv Bib

   @article{nguyen2026volatility,
     title = {Volatility-Clustering Adaptation for Financial Time Series},
     author = {Nguyen, Manh and Nguyen, Minh Hoang and Nguyen, Huu Hiep and Do, Van Dai and Le, Hung},
     year = {2026},
   }

8. SPaCe (ACL-26)

   

   SPaCe: Unlocking Sample-Efficient Large Language Models Training With Self-Pace Curriculum Learning

   Dai Do, Manh Nguyen, Svetha Venkatesh, and 1 more author

   The 64th Annual Meeting of the Association for Computational Linguistics, 2026

   TL;DR: A self-paced curriculum: clusters training data by difficulty/semantics, then a bandit allocates samples by solve rate; matches SOTA accuracy with up to 100x fewer samples.

   arXiv Bib

   @inproceedings{do-etal-2026-space,
     title     = {{SPaCe}: Unlocking Sample-Efficient Large Language Models Training With Self-Pace Curriculum Learning},
     author    = {Do, Dai and Nguyen, Manh and Venkatesh, Svetha and Le, Hung},
     editor    = {Liakata, Maria and Moreira, Viviane P. and Zhang, Jiajun and Jurgens, David},
     booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
     month     = jul,
     year      = {2026},
     address   = {San Diego, California, United States},
     publisher = {Association for Computational Linguistics},
     pages     = {3480--3507},
     doi       = {10.18653/v1/2026.findings-acl.171},
     url       = {https://aclanthology.org/2026.findings-acl.171/},
     isbn      = {979-8-89176-395-1}
   }

2025

1. GRAD

   

   GRAD: Graph-Retrieved Adaptive Decoding for Hallucination Mitigation

   Manh Nguyen, Sunil Gupta, Dai Do, and 1 more author

   2025

   TL;DR: Blends a small retrieved token graph’s evidence with the model’s own predictions at decoding time; up to +9.7% accuracy, -8.6% hallucination vs. greedy decoding.

   arXiv Bib

   @article{nguyen2025grad,
     title={GRAD: Graph-Retrieved Adaptive Decoding for Hallucination Mitigation},
     author={Nguyen, Manh and Gupta, Sunil and Do, Dai and Le, Hung},
     journal={arXiv preprint arXiv:2511.03900},
     year={2025}
   }

2. UGCS

   

   Uncertainty-Guided Checkpoint Selection for Reinforcement Finetuning of Large Language Models

   Manh Nguyen, Dung Nguyen, Dai Do, and 2 more authors

   2025

   TL;DR: Ranks RL-finetuning checkpoints by reward on the hardest, highest-uncertainty samples instead of overall validation score; more reliable than training/validation-based selection.

   arXiv Bib Code

   @article{nguyen2025uncertainty,
     title = {Uncertainty-Guided Checkpoint Selection for Reinforcement Finetuning of Large Language Models},
     author = {Nguyen, Manh and Nguyen, Dung and Do, Dai and Venkatesh, Svetha and Le, Hung},
     year = {2025},
   }

3. JoMA-25

   Topic classification of vietnamese product reviews in e-commerce using PhoBERT

   Tuan Duy Nguyen, Duc Minh Nguyen, Manh Nguyen, and 1 more author

   Journal of Marketing Analytics, 2025

   TL;DR: PhoBERT embeddings + logistic regression classify 17K Shopee reviews into 5 topics (quality, service, price, shipping, packaging); F1 = 0.96, Hamming loss = 0.022.

   DOI Bib PDF

   @article{nguyen2025topic,
     title={Topic classification of vietnamese product reviews in e-commerce using PhoBERT},
     author={Nguyen, Tuan Duy and Nguyen, Duc Minh and Nguyen, Huu Manh and Nguyen, Thi Quynh Giang},
     journal={Journal of Marketing Analytics},
     pages={1--15},
     year={2025},
     publisher={Springer}
   }

4. RDS

   

   Distance Is All You Need: Radial Dispersion for Uncertainty Estimation in Large Language Models

   Manh Nguyen, Sunil Gupta, and Hung Le

   2025

   TL;DR: A training-free uncertainty score: embeds sampled generations on a hypersphere and measures their spread from the mean; beats 9 SOTA baselines on hallucination detection.

   arXiv Bib Code

   @article{nguyen2025distance,
     title = {Distance Is All You Need: Radial Dispersion for Uncertainty Estimation in Large Language Models},
     author = {Nguyen, Manh and Gupta, Sunil and Le, Hung},
     year = {2025},
   }

5. preprint

   

   Towards Robust Fact-Checking: A Multi-Agent System with Advanced Evidence Retrieval

   Tam Trinh, Manh Nguyen, and Truong-Son Hy

   2025

   TL;DR: Four cooperating agents (claim decomposition, query generation, evidence retrieval, verdict) for automated fact-checking; +12.3% Macro F1 over baselines on FEVEROUS/HOVER/SciFact.

   arXiv Bib

   @article{trinh2025towards,
     title = {Towards Robust Fact-Checking: A Multi-Agent System with Advanced Evidence Retrieval},
     author = {Trinh, Tam and Nguyen, Manh and Hy, Truong-Son},
     year = {2025},
   }

2019

1. CinC2019

   

   Representation Learning for Early Sepsis Prediction

   Luan Tran, Manh Nguyen, and Cyrus Shahabi

   Computing in Cardiology, Apr 2019

   TL;DR: AEC-Net: an LSTM autoencoder (reconstruction loss) plus a classifier head for early sepsis prediction; scored 0.284 utility on the hidden PhysioNet/CinC 2019 test set (24th place).

   DOI Bib PDF

   @inproceedings{9005565,
     author={Tran, Luan and Nguyen, Manh and Shahabi, Cyrus},
     booktitle={2019 Computing in Cardiology (CinC)}, 
     title={Representation Learning for Early Sepsis Prediction}, 
     year={2019},
     volume={},
     number={},
     pages={1-4},
     keywords={Neural networks;Decision trees;Feature extraction;Radio frequency;Decoding;Forestry;Gold},
     doi={10.22489/CinC.2019.021}
   }