Research question
When labels are scarce, which multimodal gains survive changes in validation, operating points and data artifacts?
See the evidence ↗Research
Research questions, manuscripts and public materials.
Two first-author submissions · not yet accepted
TFT–GNN for energy–water climate riskAuditing fusion gains in low-resource learningStart here: questions, first-author work, evidence and proposed study ↗
Research focus: reliable multimodal learning under limited labels. The fake-news study motivates the evaluation question; climate-risk modeling is a separate application.
Research interests and applications
- Multimodal learning
Combining visual and textual representations for document and news understanding.
- Evaluation with limited labels
Validation sensitivity, data artifacts, honest splits and error analysis.
- Intelligent information systems
Retrieval, human-reviewed workflows, and reliable interfaces for education and services.
Research overview: existing work and proposed questions
RESEARCH AT A GLANCE
Current work and proposed direction
From low-resource news classification to reliable emotion recognition.
Evidence in hand
First-author revised manuscript: 344 archived runs studied. Public post-hoc analysis is available; 216 operating-point records were recomputed on 2 Nov. Submitted to Neurocomputing.
Check code and scope ↗Doctoral next step
Proposed work: budget-aware adaptation, evidence-aware prediction and controlled continual updates for emotion recognition.
Read the proposed questions ↗Where do fusion gains come from in low-resource multimodal fake news detection? Operating points, archive artifacts and frozen CLIP baselines
Reanalyzes 344 archived runs to separate fusion effects, operating-point changes and archive artifacts. FCP is the fusion model under study, not a newly proposed architecture.
Three takeaways and the evaluation boundary
01
Compare the operating point
Some low-shot GossipCop Macro-F1 advantages track a change in the decision operating point; a higher score alone does not establish a stronger representation.
02
Audit the data
Archive markers and near-duplicate content can confound evaluation on recollected news datasets.
03
Account for the full budget
The manuscript limits training labels but uses a fully labeled validation set. The proposed work will count selection and calibration labels too.
Scope: two English recollected news datasets. The public package supports post-hoc metrics from saved predictions; original training and split reconstruction need additional inputs. Hard labels cannot recover AUROC or score calibration.
2 Nov check: 216 result rows matched the published reference; this is an artifact check, not new training.
Hybrid AI Framework for Multi-Hazard Climate Risk Governance in Renewable Energy-Water Nexus Systems
Combines Temporal Fusion Transformer (TFT) and GNN-based Climate Risk Propagation (GNN-CRP) to study temporal risk patterns and interregional cascading risks in renewable energy–water systems across 30 Chinese provinces, 2005–2024.
What the abstract reports
Reported findings concern regional exposure differences, infrastructure-linked cascading risks and associations between governance measures and resilience. This overview is based on the author-provided abstract; numerical results and evaluation protocols are not supplied.
Validation Sensitivity in Limited-Label News Classification
The September 2026 CV records this as submitted. Author order, current status and an official proceedings or DOI link still require confirmation.
Research summary
Studies validation sensitivity in news classification with limited labels.
Design and Implementation of a Vision-Language Multimodal Behavior Prediction Model
The September 2026 CV records first authorship and acceptance. This page has no linked official proceedings citation or DOI, and the status needs a current check.
Research summary
Encodes visual and language inputs with VideoMAE and DistilBERT, combining bidirectional cross-attention, gated residuals, text dropout and LSTM temporal modeling to predict behavior classes and trajectory trends.
This method summary is not a reported metric.
Dental Deformities Prediction Using Bayesian Networks among Adolescents
Research summary
Master's research using R and Bayesian networks to study conditional relationships and dental deformities prediction among adolescents.
Five selected records; all seven appear in the academic CV. Climate Risk Management follows the abstract and first-author status confirmed on 27 October 2026. Neurocomputing follows the revised manuscript and 24 October update; other entries follow the supplied CV.