Research / Technology & Engineering
Technology & Engineering
We build AI systems that stay reliable when their inputs are uncertain: perception that makes mistakes, memories that must be searched with vague cues, and cities observed through sparse sensors.
Technology research at OSiV is carried out with our partner group ViSO. The common thread is structure and calibrated uncertainty: instead of trusting one prediction, our systems keep structured records with confidences and let reliable evidence outweigh unreliable evidence.
We work on video understanding, memory for AI agents and spatio-temporal learning, and evaluate on public benchmarks with controlled noise so that robustness can be measured, not assumed.
Research themes

Video event understanding under perceptual uncertainty
How can a retrieval system keep finding events when detectors, trackers and action classifiers make mistakes?
Long videos are summarised as structured events (time, location, actor, action). We treat the correctness of each attribute as a latent variable and calibrate its confidence, so one misrecognised cue no longer discards an event.

Structured memory for AI agents
What makes structured episodic retrieval robust when the query cues themselves are noisy?
LLM agents need to recall past episodes from partial and sometimes wrong cues. We study, under controlled noise, how confidence-aware matching across several memory axes compares with strict matching and embedding search.

City-scale spatio-temporal learning
How can sparse road sensors and dense vehicle trajectories be combined to predict traffic across a whole city?
Graph neural networks fuse loop-detector data with taxi GPS trajectories; pattern-driven pseudo-labels fill in roads that have no sensor.
Publications
-
Authors withheld during reviewUnder review, 2026 -
Probabilistic Structured Event Modeling for Video Event Recognition under Perceptual Uncertainty
Nhan Thanh Ho, Anh Dong Nguyen, Thuan Duc Tran, Thang Van Nguyen, Bui Anh Thinh, Nguyen Bao Tran, Long Nhat HoAsian Conference on Computer Vision (ACCV) 2026ViSO -
24H-GNN: A Pattern-Driven Pseudo-Label Fusion Graph Network for City-Wide Traffic Flow Prediction
Thanh Nhan Ho, Dong Anh Nguyen, Minh Tai Pham Nguyen, Minh Khue Phan Tran, Hoang Minh Nguyen, Nguyen Bao Tran, Trong Nhan LeInternational Conference on Intelligent Systems and Networks (ICISN) 2026
Topics
See also: Economics, Culture & Society