Institution profile

Hanoi University of Science and Technology

Academic institutionasia · vn
Official website
Research library221linked papers
Opportunities0open roles
Selected work

Representative Papers

Boosting Offline Optimizers with Surrogate Sensitivity

Mar 06, 2025International Conference on Machine Learning

Offline optimization of expensive black-box functions in materials engineering suffers from poor robustness due to the high sensitivity of surrogate models to parameter perturbations. Method: We propose, for the first time, an optimizable surrogate sensitivity metric and design a sensitivity-aware regularization method orthogonal to existing frameworks. This approach integrates gradient-based sensitivity analysis with deep-learning-based surrogate modeling and is compatible with mainstream paradigms such as offline Bayesian optimization. Contribution/Results: Evaluated on multiple materials design benchmarks, our method significantly improves optimization success rate (average gain of +23.6%) and solution quality (objective value improvement up to 17.4%). Empirical results demonstrate that explicit sensitivity control delivers critical performance gains for offline optimization of expensive black-box functions in materials engineering.

2 citations1 influentialRead paper

ViQA-COVID: COVID-19 Machine Reading Comprehension Dataset for Vietnamese

Apr 21, 2025arXiv.org

To address the scarcity of Vietnamese-language machine reading comprehension (MRC) resources and dedicated NLP benchmarks for pandemic-related text in low-resource languages, this work introduces ViQA-COVID—the first Vietnamese multi-span, multi-paragraph MRC dataset focused on COVID-19. Constructed manually from authentic pandemic documents, it comprises thousands of high-quality question-answer pairs annotated under a fine-grained multi-span schema, enabling robust fine-tuning and evaluation of mainstream models (e.g., PhoBERT). Its key contributions are twofold: (1) it pioneers multi-span answer extraction for Vietnamese MRC, and (2) it fills a critical gap by providing the first domain-specific MRC benchmark for pandemic-related text in a low-resource language. Empirical results demonstrate that ViQA-COVID substantially improves model performance on Vietnamese pandemic text understanding and has already enabled multiple follow-up studies in health-focused NLP.

1 citations1 influentialRead paper

Federated Prompt-Tuning with Heterogeneous and Incomplete Multimodal Client Data

Feb 06, 2026

This work addresses the challenge of semantic misalignment in federated learning caused by heterogeneous multimodal client data and missing input-level features. To tackle this issue, the paper proposes the first federated multimodal prompt tuning framework, which enables collaborative optimization and effective fusion of prompt instructions across clients and modalities. The approach integrates client-specific prompt tuning with a server-side semantic-aware aggregation mechanism, establishing a novel paradigm that supports prompt alignment and aggregation under heterogeneous missing-data patterns. By innovatively combining federated learning with multimodal prompt tuning, the method achieves significant performance gains over state-of-the-art baselines across multiple multimodal benchmark datasets, demonstrating its effectiveness and robustness.

1 citationsRead paper

Incorporating Surrogate Gradient Norm to Improve Offline Optimization Techniques

Mar 06, 2025Neural Information Processing Systems

In offline optimization, surrogate models suffer from poor calibration in out-of-distribution regions; existing conditional methods exhibit weak generalization and strong model dependency. This paper proposes a model-agnostic gradient norm regularization that explicitly constrains the local sharpness of surrogate models during training. We are the first to extend sharpness-based generalization theory—from prediction loss to the gradient level—establishing a theoretical bound linking training-set gradient sharpness to worst-case gradient sharpness on unseen data. The proposed regularization is architecture-agnostic and seamlessly integrates into arbitrary surrogate models (e.g., Gaussian processes, neural networks) without structural modification. Empirical evaluation on multi-objective black-box optimization tasks demonstrates an average performance improvement of 9.6%, with significant gains in generalization and robustness. The implementation is publicly available.

1 citationsRead paper
Recent publications

Latest Papers