Scholarship

Publications

Peer-reviewed research on trustworthy AI, human-in-the-loop decision systems, and NLP for low-resource languages.

Published

PUBLISHED · 2026

Proceedings of SemEval-2026 (ACL) · pp. 1632–1640 · San Diego, USA

DeepSemantics at SemEval-2026 Task 9: Label-Wise Optimization with Adaptive Focal Loss for Polarization Manifestation Identification

TIAO, E., Edou, J.R., Gohouedé, M.A.L.

Impact: Multilingual system addressing class imbalance in low-resource language NLP. Ranked 5th on Hausa track, 14th on English among 50+ teams.

End-to-end transformer pipeline for detecting fine-grained polarization manifestations in English and Hausa social media. Uses Adaptive Focal Loss and label-wise threshold optimization to handle severe class imbalance in multilingual settings.

Code ↗ ORCID ↗

Accepted for publication

ACCEPTED · 2026

Deep Learning Indaba 2026 / IJCAI Special Volume

Data-Driven Lifelines: An Artificial Intelligence and Econometric Early-Warning System for Small and Medium-Sized Enterprise Survival in Cameroon

Konlack Lonlack, G., TIAO, E.

Impact: Deployed as interactive policy dashboard at Cameroon's National Institute of Statistics. Identifies high-failure-risk enterprises to enable targeted government support.

Hybrid econometric-ML pipeline integrating Lasso feature selection with ensemble methods (CatBoost, Random Survival Forest) on national enterprise census data. Generates firm-level survival curves and risk scores for policy intervention.

ACCEPTED · 2026

Deep Learning Indaba 2026 / IJCAI Special Volume

FloodTriage: A Decision-Theoretic Tri-Fold Deferral Policy Towards Human-in-the-Loop Streamflow Forecasting

Gohouedé, L.C., Gohouedé, M.A.L., TIAO, E., Edou, J.R.

Impact: Reduces false flood alarms while maintaining 100% miss detection. Routes only 25% of predictions to humans for review in resource-constrained settings.

Human-AI collaboration framework for flood forecasting on the Ouémé River (Cameroon). LSTM deep ensemble governed by tri-fold deferral gateway that decomposes decisions into epistemic, aleatoric, and Knightian uncertainty signals.

Under review

UNDER REVIEW · 2026

TrustAI Workshop · Deep Learning Indaba 2026 · Proceedings of Machine Learning Research

Trustworthy Human-AI Collaboration for SME Survival: A Cost-Aware and Uncertainty-Guided Deferral Framework

TIAO, E., Gohouedé, M.A.L., Gohouedé, L.C., Edou, J.R., Mwanda, V.

Key innovation: Cost-sensitive routing that decomposes model uncertainty to decide automation vs. deferral, balancing expected cost, fairness criteria, and undetected-failure risk.

Framework for human-AI collaboration in high-stakes predictions by decomposing uncertainty into model confidence, data scarcity signals, and unknowns. Routes decisions based on business cost and fairness constraints.

Master's thesis research (Sep 2025–Jul 2026) — Probabilistic forecasting of electricity consumption and solar PV production with explainable uncertainty attribution. Methods: quantile regression, conformal prediction, SHAP-based feature attribution. Deployed for use in production energy systems.