Research
Research projects at the Institute
Vietnamese Natural Language Processing
Vietnamese has particular characteristics in vocabulary, tone marks and expression that make many language tools developed for English less effective. This research direction aims to build and evaluate natural language processing methods suited to Vietnamese, prioritizing applications close to education and services. Objectives • Study and test large language models on Vietnamese data • Aim at question answering, summarization and text classification tasks in Vietnamese • Propose quality evaluation methods that fit the characteristics of Vietnamese Approach The research team intends to combine the lawful collection and normalization of text data, fine-tuning of existing models, and evaluation on each specific task. Issues of data quality, bias and personal information protection are considered from the outset. Expected Outputs • Datasets and evaluation criteria to support research • Sample models and tools supporting teaching and information lookup • Technical reports and guidance materials for the community
AI-Assisted Learning Analytics
Data generated from online and in-class learning can help lecturers understand more clearly where learners are struggling. This research direction aims to apply data analysis and machine learning to support personalized learning, while ensuring privacy and transparency. Objectives • Understand students' learning patterns from learning data collected with consent • Aim at early warning of dropout risk or falling behind schedule • Suggest learning materials and activities suited to different groups of learners Approach The research team intends to build a controlled data collection process, anonymize identifying information, and develop explainable models so that lecturers can understand and verify the results. Model outputs are treated as reference information supporting human decisions, not as a replacement for lecturers' judgment. Expected Outputs • Visual dashboards for lecturers and academic advisors • An ethical framework and rules for the use of learning data • Guidance materials for application in courses
Computer Vision for Quality Inspection
Quality inspection by the human eye is time-consuming and difficult to keep consistent when production volumes are large. This research direction aims to apply computer vision to help detect surface defects and shape deviations in products, focusing on practical, affordable solutions for small and medium enterprises. Objectives • Study anomaly detection methods when defect data is scarce • Aim at lightweight systems that can run on common hardware • Assess model reliability under changing lighting and environmental conditions Approach The research team intends to start by building an image collection and labeling process, then test object detection and anomaly detection models in a laboratory setting. Deployment at production sites will be considered only once appropriate agreements are in place and data safety is ensured. Expected Outputs • A reference process for building image-based inspection systems • Sample software for teaching and practice • Technical documents sharing deployment experience
Data Platforms for Digital Transformation in Universities
Universities generate large volumes of data from teaching, research and administration, yet the data is often scattered across many systems. This research direction aims to design a unified and secure data architecture that helps institutions make evidence-based decisions. Objectives • Propose a data warehouse and data governance architecture suited to the university context • Aim at standardizing data catalogs, data quality and access rights • Support reporting and analysis for management work Approach The research team intends to survey the current state of data, build a reference data model and test integration on a small scale before expanding. Principles of security, access control and compliance with personal data regulations are mandatory requirements in the design. Expected Outputs • A reference architecture and data governance guidelines • A pilot data analytics dashboard • Documents sharing experience with other educational institutions
Trustworthy and Explainable AI
For AI to be widely used in sensitive fields such as education, finance and healthcare, systems need to be trustworthy, explainable and accountable. This research direction aims to study and propose methods that help users understand why a model produces a particular result. Objectives • Study model explanation techniques and assess how useful they are to users • Aim at detecting and reducing bias in data and models • Propose a risk checklist to use before deploying AI systems Approach The research team intends to combine literature review, experiments on public datasets, and consultation with end users to assess how understandable the explanations are. A human-centered approach is prioritized. Expected Outputs • A responsible AI assessment framework for projects within the university • Sample tools supporting fairness checks and model explanation • Training materials on responsible AI for students and lecturers
AI for Public Services and Local Digital Transformation
Online public services are becoming more common, yet citizens and officials still face difficulty in looking up information, preparing documents and handling large volumes of text. This research direction aims to explore how AI can make public services more accessible and efficient, on a foundation of transparency and personal data protection. Objectives • Survey needs and bottlenecks in the public service delivery process • Aim at a Vietnamese-language assistant for answering questions about procedures • Study support for classifying and summarizing administrative documents Approach The research team intends to proceed step by step: surveying needs, building a prototype on public data, then evaluating it together with users. All solutions are designed to support officials and citizens, and not to replace the decision-making responsibility of competent authorities. Expected Outputs • A survey report and prioritized solution proposals • A prototype information assistant for research and training purposes • Guidelines for deploying AI safely in agencies and organizations