Peer-reviewed work in Bayesian machine learning, uncertainty quantification and evidence-governed AI. These methods shape how TERA products handle evidence, risk and doubt.
Each entry gives the journal, the authors, the core contribution and a direct link to the paper. View counts and impact claims are left out on purpose.
Machine Learning and Knowledge Extraction, 8(6):151Open access
Lebede Ngartera, Saralees Nadarajah, Rodoumta Koina, Youssou Gningue
A Bayesian framework that measures uncertainty over retrieved evidence and carries it into generation, so a system can qualify, defer or abstain when support is insufficient.
Journal of Risk and Financial Management, 19(3):173
Lebede Ngartera, Saralees Nadarajah, Rodoumta Koina, Youssou Gningue
Links calibrated quantile forecasts to constrained inventory decisions, and judges them by tail risk (Conditional Value-at-Risk) rather than average performance alone.
Frontiers in Artificial Intelligence, Volume 8
Lebede Ngartera, Saralees Nadarajah, Rodoumta Koina
Builds uncertainty into retrieval with Monte Carlo Dropout and a risk-sensitive score that balances relevance against uncertainty in financial question answering.
Frontiers in Built Environment, Volume 11
Lebede Ngartera, Saralees Nadarajah
Uses Bayesian neural networks in autonomous navigation so that predictive uncertainty triggers more cautious behavior under noisy sensing, bad weather and unfamiliar traffic.
IEEE Access, Volume 13
Lebede Ngartera, Mahamat Ali Issaka, Saralees Nadarajah
Compares interpretable Naïve Bayes models with hybrid deep-learning and ensemble methods on email phishing and highly imbalanced card-fraud data.
Machine Learning and Knowledge Extraction, 6(4)
Lebede Ngartera, Mahamat Ali Issaka, Saralees Nadarajah
Three clinical cases, including diabetes outcomes and early Alzheimer’s detection, showing how predictive accuracy and uncertainty can be weighed together.
Open Journal of Optimization, Volume 13
Lebede Ngartera, Coumba Diallo
Benchmarks gradient descent, momentum and adaptive methods on a hard non-convex function, comparing convergence speed and stability.
A deterministic method that interprets air-quality episodes only as far as independent evidence agrees, and abstains otherwise. First results cover 116 episodes in Delaware County, Pennsylvania, from 2022 to 2025.
Related product: TEIEvery project follows the TERA standard: Trustworthiness, Efficiency, Reliability and Accountability. Read our philosophy.
We work with researchers, universities and public-interest organizations on uncertainty-aware systems, evidence governance and evaluation.
