Updates
News
Paper acceptances, awards, talks and other updates from the lab.
2026
Swin-AQI accepted at ACIVS 2026
Swin-AQI estimates PM2.5 and PM10 concentrations from street-level photographs using a depth-aware Swin Transformer. The paper has been accepted at ACIVS 2026 in Okinawa, Japan.
Paper on fuzzy risk-scaled robot recovery accepted at ACIVS 2026
Instead of an all-or-nothing emergency stop, the system reads prediction error in latent space as a continuous risk signal and scales trajectory corrections accordingly. Accepted at ACIVS 2026.
Paper on multispectral image retrieval accepted at ACIVS 2026
A test-time adaptation method lets a frozen CLIP backbone handle multispectral satellite imagery with no Earth-observation pretraining. Accepted at ACIVS 2026.
GraphMemShield accepted at CIKM 2026 (CORE Rank A)
A framework for auditing cross-session data leakage in the knowledge-graph memories behind AI assistants, accepted into the CIKM 2026 Full Research Track at a 27% acceptance rate.
Paper on examination room optimization accepted at SOMET 2026
A case study on real data from Ho Chi Minh City University of Technology, combining a greedy heuristic with integer linear programming to cut the number of exam rooms that must be opened.
Paper on LLM agent reliability accepted at ISSRE 2026 (CORE Rank A)
TraceToChain fits LLM agent execution traces to an absorbing Markov chain, showing that pass@k, pass^k and the reliability decay curve are projections of one success-time distribution.
Outstanding Student Paper Award at FLINS-ISKE 2026
A paper on forecasting and inventory optimization for ATM cash management in Vietnam won the Outstanding Student Paper Award at FLINS-ISKE 2026 in Sydney, Australia.
Survey on industrial visual anomaly detection in robotics at IEA/AIE 2026
A survey of deep learning methods for industrial visual anomaly detection, focused on their role in safety monitoring for collaborative robots.
Survey on event extraction with large language models at IEA/AIE 2026
An up-to-date overview of LLM-based event extraction, with unusual attention to multilingual data resources — Vietnamese among them.
Paper on LLM agents in static analysis nominated for Best Paper at IEA/AIE 2026
A four-dimensional taxonomy for LLM agents in static analysis, nominated for the Best Paper Award at IEA/AIE 2026.
Best Paper Award at IEA/AIE 2026 for research on explainable AI in software engineering
A systematic mapping study of explainable AI across the software development life cycle won the Best Paper Award at IEA/AIE 2026.
Distributional Program Analysis accepted at ASE 2026 (NIER Track, CORE Rank A*)
Rather than analysing a single LLM-generated program, the authors analyse many samples at once and treat an invariant's survival rate as cross-sample evidence. Accepted into the ASE 2026 NIER Track at 25%.
GraphLedger accepted at ASE 2026 (Research Track, CORE Rank A*)
GraphLedger detects vulnerabilities at repository scale by pairing graph-based compression with an assumption ledger. Accepted into the ASE 2026 Research Track (115/1236).
Paper on Vietnamese event extraction published in IJIC
A small language model framework augmented with external knowledge, aimed at Vietnamese event extraction under tight computational budgets.
VSLIM published at ACIIDS 2026
A joint model for Vietnamese multi-intent detection and slot filling, using a biaffine classifier to capture the link between intents and their slots directly.
SBV-LawGraph presented as an oral paper at ACIIDS 2026
A hybrid RAG framework integrating a legal knowledge graph for State Bank of Vietnam documents, selected for oral presentation at ACIIDS 2026.