Faculty of Computer Science and Engineering · HCMUT, VNU-HCM
Reasoning in Artificial Intelligence and Software Engineering
We develop methods for building AI systems that can reason, verify what they produce, and know when they are uncertain.
Best Paper Award — IEA/AIE 2026
From Black-Box to Glass-Box: Explainable AI for Applied Intelligent Software Systems
Outstanding Student Paper Award — FLINS-ISKE 2026
Uncertainty-Aware Forecasting and Inventory Optimization for ATM Cash Management in Vietnam
VNU350 Outstanding Young Scientist, 2025
Lab head Xuan-Bach Le was recruited to HCMUT through VNU-HCM's VNU350 Programme.
Research areas
All research →Our work sits at the intersection of formal methods, large language models, and software reliability: we use logic and verification to give evidence for what learned systems claim, and we use learned systems to reach problems that classical verification cannot.
Verified AI
Formal reasoning · program synthesis · verificationLogics, decision procedures and proof techniques that certify what a program — or a model — is claimed to do, from separation logic and fractional permissions to LLM-driven program analysis.
11 papers →02Auditable AI Agents
LLM agents · RAG · reliability · securityMaking LLM agents and retrieval pipelines accountable: auditing their traces, defending their memories and retrieval corpora, and deciding which of their outputs have earned the right to be trusted.
14 papers →03Safe Autonomous Systems
RL · robotics · anomaly detection · risk-aware controlAutonomy that respects a specification: reinforcement learning under temporal-logic objectives, and detecting and recovering from anomalies before an autonomous system acts on them.
3 papers →04Quantum Reasoning
Quantum programs · quantum logic · post-quantum securityReasoning about quantum software when running it is not enough: separation logic for quantum state, retrieval-grounded repair of quantum programs, and the control-plane decisions that make services quantum-safe.
3 papers →Latest news
All news →Invited talks
Software Engineering in the Era of LLMs: From Formal Verification to Auditable Agents
Invited talk, ICEBA 2026 — 7th International Conference on Engineering, Physics, MEMS-Biosensors and Applications. Vung Tau, Vietnam, November 2026. Slides
Beyond General-Purpose LLMs: Building Trustworthy, Domain-Specific AI for Finance and Law in Vietnam
Invited talk, Van Lang University — Workshop on AI in Finance and Law. Ho Chi Minh City, Vietnam, May 2026. Slides
Selected recent publications
All 84 publications →Stability Across Environments, Not Resamples: Noise-Corrected Symbolic Pattern Mining
WSDM 2027 (Main Conference; acceptance rate: 15.8%), Hong Kong · February 2027
A Selection-Aware Audit of Learned Stopping in Short-Horizon Iterative Retrieval
WSDM 2027 (Main Conference; acceptance rate: 15.8%), Hong Kong · February 2027
EvidenceGraph: Measuring and Controlling Repeated Evidence in Retrieval-Augmented Verification
WSDM 2027 (Main Conference; acceptance rate: 15.8%), Hong Kong · February 2027
When Do Propagation Models Beat Cascade Summaries? A Released-Signal Audit of Text-Free Misinformation Detection
WSDM 2027 (Findings; acceptance rate: 12.6%), Hong Kong · February 2027
Anchor the Head: Alignment Drift in Open-Vocabulary Segmentation
WACV 2027 (Applications Track, CORE Rank A), Orlando, Florida, USA · January 2027
SegMaFormer: A Hybrid State-Space and Transformer Model for Efficient 3D Medical Image Segmentation
WACV 2027 (Applications Track, CORE Rank A), Orlando, Florida, USA · January 2027
Work with us
We are looking for motivated students who want to do careful, honest research on reliable AI and software engineering — work that is worth publishing.