LVT
Le Van Truong
02 · Scholarly Publications & Technical ManuscriptsORCID: 0009-0008-7015-7392

Publications & Preprints

Complete scholarly record featuring first-author research papers in international conferences and workshops (ACM ICMR LSC '26, KES '26, ICMV '25, FDSE '25, FJCAI '26), arXiv preprints, active under-review manuscripts, and collaborative lab & national symposium works.

First-Author Manuscripts
7
6 Published/Accepted + 1 arXiv
Published & Accepted
6
ACM, Elsevier, Springer, SPIE
Preprints & Under Review
2
arXiv:2604.01039 & ViProDID
Lab & Symposium Works
7
ICMR, ISMAR & National Proc.
Status:Year:
Domain:
Showing 8 of 8 primary research manuscriptsAuthor order & venue status verified
01PublishedACM ICMR 2026 · LSC '26· ACM Digital Library · Scopus (Rank B)
First Author2026

ChronoSeek: Temporal-First Agentic RAG with Event Knowledge Graph for Lifelog Retrieval

Van-Truong Le, Minh-Triet Tran, Viet-Tham Huynh

The 9th Annual ACM Workshop on the Lifelog Search Challenge (LSC '26), co-located with ACM ICMR 2026, Amsterdam, Netherlands

Quick Summary (What + Result)

Designed a temporal-first Event Knowledge Graph (EKG) schema in Neo4j encoding NEXT, SAME_PLACE, and SIMILAR_ACTIVITY edges over segmented lifelog events.

LSC '26 Task 3 (QA) Score
1000.0 / 1000.0 (Rank #1)
Highest score among all 15 participating teams globally
Zero-Shot Event-Aware MRR
0.607
10.8× improvement over static CLIP (0.056) and SigLIP (0.00)
Oracle Event-Aware MRR
1.000
Validated on both 7K-image dev set and 118K-image 3-month subset
PDF (Direct)Paper PageSystemDOILab Repo
Architecture / Figure
High-level architecture of ChronoSeek showing Offline Indexing and Online Retrieval & Reasoning

Figure 1: High-level architecture of ChronoSeek combining offline Event Knowledge Graph + Milvus indexing with online LangGraph Agentic Orchestration and Fast-Path fallback.

02PublishedKES 2026 · Procedia CS· Elsevier Procedia Computer Science · Scopus (Rank B)
First Author2026

Enhancing Multilingual Fact-Checking Retrieval with Ensemble Bi-Encoders and Architecture-Aware Hard Negative Reranking

Van-Truong Le, Duy-Minh Nguyen-Tran, Tung Le, Huy Tien Nguyen

30th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES 2026), Procedia Computer Science 292 (2026) 1845–1853

Quick Summary (What + Result)

Proposed a two-stage multilingual fact-checking pipeline combining a 5-model BiLSTM-Attention E5-Large ensemble with a BGE-M3 Cross-Encoder reranker.

Cross-Lingual S@10
0.8833
+8.0% over Stage-1 Dense Ensemble (0.8033) & +2.46% above 1st-place PINGAN AI (0.8587)
Overall S@10
0.9163
Monolingual S@10 = 0.9215; reduces mono-to-cross-lingual gap from 9.40% to 3.82%
Negative Mining Ablation
0.9163 vs. 0.2015
Dense-mined hard negatives (0.9163) vastly outperform Random (0.8049) and BM25-mined (0.2015)
PDF (Direct)Paper PageSystem
Architecture / Figure
Two-stage multilingual fact-checking retrieval architecture with 5x Bi-Encoder Ensemble and BGE-M3 Cross-Encoder

Figure 1: Overview of the two-stage framework: 5× E5-Large + BiLSTM-Attention Ensemble (Stage 1) and Dense Hard-Negative BGE-M3 Reranker (Stage 2).

03PreprintarXiv:2609.11166 · Zenodo· arXiv (cs.LG) · Zenodo Permanent Archive
First Author2026

When does a spectral prior help graph learning? Connectivity-loss estimation under road-network disruptions

Van-Truong Le

arXiv Preprint (cs.LG) & Zenodo Archival Release (v1.0.0)

Quick Summary (What + Result)

Formulated a bounded residual spectral GNN estimator combining closed-form Fiedler edge sensitivity with learned nonlinear corrections for simultaneous multi-edge deletions.

Zero-Shot OSM Spatial MAE
0.096 → 0.057
Residual GCN paired gain of 0.0391 [95% CI: 0.0151, 0.0662] on expanded 13-area panel
Targeted Disruption Gain
+0.0257 (Zero-Shot) / +0.0622 (LOCO)
Statistically supported improvement under edge-betweenness targeted attacks
20,000-Node Screening Speedup
56× Faster (23.17 ms vs. 1,302.4 ms)
Amortized over 1,000 screened disruption scenarios per intact eigensolver setup
PDF (Direct)arXivZenodoPaper PageCodeSystem
Architecture / Figure
ResiliRoad workflow diagram showing base spectrum, fast prior, and GNN residual correction

Figure 1: ResiliRoad workflow separating intact spectral prior computation from GNN residual correction.

04AcceptedICMV 2026· SPIE / ICMV Proceedings · Scopus (Rank C)
First Author2026

SKIA: Spatial-Knowledge Informed Attention for Robust Remote Photoplethysmography under Lower-Face Motion Artifacts

Van-Truong Le, Viet-Tham Huynh, Minh-Triet Tran

19th International Conference on Machine Vision (ICMV 2026)

Quick Summary (What + Result)

Characterized lower-face motion contamination in rPPG and proved that a static anatomical prior (α = 0.7) significantly improves CHROM across 88.1% of subjects (p = 2.0 × 10⁻⁷).

SKIA-Net Full Performance
ρ = 0.7522 · HR MAE = 3.73 BPM
Outperforms PhysNet-lite (ρ = 0.2350), SKIA Raw (8.99 BPM), and SKIA Masked (7.55 BPM)
Masked CHROM Gain
ρ: 0.3507 → 0.4166 · MAE: 33.50 → 23.44 BPM
88.1% subject winrate across 42 subjects (Wilcoxon p = 2.0 × 10⁻⁷)
Static vs. Dynamic Optical Flow
0.4166 vs. 0.1534 Pearson
Proves temporally invariant anatomical priors avoid frame-wise color sampling drift
PDF (Direct)Paper PageCodeSystem
Architecture / Figure
Overview of the SKIA multi-stream 3D-CNN framework for remote photoplethysmography

Figure 1: Overview of the SKIA framework combining raw frames, static lower-face suppressed frames, and temporal residuals.

05AcceptedFDSE 2026 · Springer CCIS· Springer CCIS · Scopus
First Author2026

FLeX: Fingerprint Leakage Evaluation across Super-Resolution and Cross-Domain Matching

Van-Truong Le, Le-Khanh Nguyen, Tran-Chau Vo Han, Hieu-Hoc Tran Minh, Minh-Triet Tran

13th International Conference on Future Data and Security Engineering (FDSE 2026), Springer CCIS

Quick Summary (What + Result)

Introduced FLeX, a three-level biometric leakage protocol separating visual fidelity, template extractability, and downstream cross-domain identity matching.

Reconstruction Fidelity
Bicubic 25.333 dB vs. SwinIR 24.696 dB PSNR
Bicubic outperforms generic SwinIR across PSNR, SSIM (0.518 vs 0.485), and LPIPS (0.513 vs 0.572)
Minutiae Extractability Surge
51.19 (Raw) → 197.50 (SwinIR) → 421.94 (Clean+Enh)
3.86× to 8.24× more detected minutiae compared to raw contactless images
Cross-Domain Matching EER
0.4998 (SwinIR) · 0.5008 (Raw)
Near-random EER ≈ 0.5 across all contactless probes (vs. 0.0638 same-domain livescan sanity check)
PDF (Direct)Paper PageSystemLab Repo
Architecture / Figure
Full pipeline of the FLeX biometric leakage assessment framework

Figure 1: Three-level FLeX evaluation pipeline linking super-resolution quality, MINDTCT minutiae extraction, and Bozorth3 cross-domain matching.

07AcceptedFJCAI 2026 · arXiv:2604.16234· National AI Conference Proceedings · arXiv:2604.16234
First Author2026

A Two-Stage, Object-Centric Deep Learning Framework for Robust Exam Cheating Detection

Van-Truong Le, Le-Khanh Nguyen, Trong-Doanh Nguyen

FISU Joint Conference on Artificial Intelligence (FJCAI 2026), Vietnam & arXiv (cs.CV)

Quick Summary (What + Result)

Proposed a two-stage object-centric exam behavior analysis pipeline pairing YOLOv8n ROI extraction with RexNet-150 classification.

Test Set Accuracy
95.16%
+16.64% absolute improvement over Full-Frame RexNet-150 baseline (78.52%)
Minority-Class ('cheating') F1
0.91 (Precision 0.91 · Recall 0.91)
+0.28 F1 gain over Full-Frame baseline (0.63 F1); Macro/Weighted F1 = 0.94 / 0.95
Inference Latency
13.9 ms / sample
Real-time throughput (~72 FPS) on a single RTX 3080 GPU
PDF (Direct)arXivPaper PageSystem
Architecture / Figure
Two-stage Exam Cheating Detection architecture with YOLOv8 localization and RexNet-150 classification

Figure 1: Two-stage object-centric pipeline isolating examinee ROIs via YOLOv8n before RexNet-150 behavior classification.

08Under ReviewUnder Review (2026)· Submitted Manuscript · Public Code Verified
First Author2026

ViProDID: Speaker-Disjoint Fine-Grained Vietnamese Dialect Identification across 63 Provinces via Prosodic Gating and Hierarchical Mixture-of-Experts

Van-Truong Le

Manuscript Submitted for Peer Review (Official Implementation Available)

Quick Summary (What + Result)

Developed ViProDID (`Hierarchical-DialectMoE`) for speaker-disjoint 63-province Vietnamese dialect identification.

63-Province Macro-F1
59.32 ± 0.18%
Statistically significant improvement (paired-bootstrap p < 0.001) on speaker-disjoint ViMD
3-Region Macro-F1
94.42 ± 0.15%
High macro-regional separability across North, Central, and South dialects
Paper PageCodeSystem
Collaborative Lab Outcomes & National Conference Proceedings

Additional Scholarly Works & Symposium Presentations

Collaborative outcomes from VNG – Saigon AI Hub Applied AI Lab Batch 1 and presentations in national scientific conferences and university proceedings.

01 · Collaborative Lab Outcome (Cited in CV)2026

Lumina-HT2 (Collaborative Outcome — VNG Saigon AI Hub Applied AI Lab)

ACM International Conference on Multimedia Retrieval (ACM ICMR 2026)

Outcome of VNG – Saigon AI Hub Applied AI Lab Batch 1 research cohort.

02 · Collaborative Lab Poster (Cited in CV)2026

CrisisX: Mobile VR Clinical-Communication Training with Adaptive Agents and VAD Emotion Modeling

IEEE International Symposium on Mixed and Augmented Reality (IEEE ISMAR 2026 Poster)

Evaluated with 14 healthcare-related students (UMUX-LITE 76.8/100); developed in VNG – Saigon AI Hub Applied AI Lab.

03 · University Scientific Conference Presentation2026

Tăng cường truy vấn kiểm chứng thông tin đa ngôn ngữ với kỹ thuật Ensemble Bi-Encoders và Reranking tối ưu kiến trúc

3rd HCMUS Young Scientists Conference (Hội nghị Khoa học Trẻ Trường ĐH KHTN, ĐHQG-HCM Lần 3)

Presented research on multilingual fact-checking retrieval.

04 · National Workshop Proceedings (ISBN: 978-632-031-603-8)2026

Năng lực số của cán bộ thư viện trong kỷ nguyên Trí tuệ Nhân tạo: Đánh giá thực trạng và đề xuất giải pháp

Scientific Conference Proceedings: Digital Library & AI in Modern Higher Education (HUB & Labor Publishing House)

Published in print proceedings (ISBN 978-632-031-603-8).

05 · VNU-HCM Proceedings (ISBN: 978-632-608-568-6)2025

Kiến tạo tương lai từ giảng đường: Vai trò xung kích của sinh viên và cán bộ trẻ tại ĐHQG-HCM trong hệ sinh thái đổi mới sáng tạo quốc gia

VNU-HCM Scientific Research & Innovation Conference for Students and Young Faculty

Published in conference proceedings (Nov 2025).

06 · Domestic Symposium Paper2025

So Thong Thai: A Gamified Mobile-Based Solution to Promote the 'Digital Literacy for All' Movement for the Elderly

Ho Chi Minh City Scientific Workshop on Youth in the Digital Literacy Movement

Presented June 2025.

07 · Domestic Symposium Paper2025

Building an Integrated Student Support Model in Online Education in Vietnam: From Situational Analysis to a Strategic Action Framework

Hanoi Open University Symposium on Student Experience and Support in Online Education

Presented December 2025.