Selected work

Representative Papers

Feature-Based Likelihood Ratios for Forensic Science: Combining Neural Networks with Bayesian Probability Calculus

Sep 25, 2026

This study addresses the calibration challenges inherent in feature-based likelihood ratio (LR) models within forensic science, which frequently compromise the statistical interpretation of “similarity” and “typicality.” To overcome these limitations, this work proposes a bi-level model that integrates neural network gradient descent with Bayesian probability theory to generate well-calibrated, on-the-fly feature-based LRs. By restoring the LR as a comparative measure of similarity and typicality, the proposed framework achieves end-to-end calibration without requiring post-processing. Evaluated on glass fragment LA-ICP-MS data, the method yields a 4.5-fold improvement in the log-likelihood ratio cost (C_llr) over comparable systems, significantly outperforming existing feature-based LR approaches. These results establish an interpretable and highly reliable new paradigm for forensic evidence evaluation.

0 citationsRead paper

From specific-source feature-based to common-source score-based likelihood-ratio systems: ranking the stars

Apr 24, 2026

This study addresses the evaluation and selection of source-level likelihood ratio (LR) systems for forensic evidence-to-reference comparison tasks by proposing an integrated analytical framework that balances performance and practical feasibility. The authors employ strictly proper scoring rules to quantify how effectively each system updates Bayesian prior odds and present the first systematic comparison among specific-source feature-based, common-source anchored, and unanchored score-based LR approaches. Their findings reveal that specific-source feature-based LRs achieve the highest performance but incur substantial experimental costs, whereas common-source feature-based methods offer strong discriminative power with significantly reduced implementation complexity. All LR systems substantially outperform a baseline relying solely on prior odds. This work thus provides both theoretical grounding and practical guidance for selecting LR systems in forensic practice.

0 citationsRead paper

Recent publications

Latest Papers

Feature-Based Likelihood Ratios for Forensic Science: Combining Neural Networks with Bayesian Probability Calculus

Sep 25, 2026

This study addresses the calibration challenges inherent in feature-based likelihood ratio (LR) models within forensic science, which frequently compromise the statistical interpretation of “similarity” and “typicality.” To overcome these limitations, this work proposes a bi-level model that integrates neural network gradient descent with Bayesian probability theory to generate well-calibrated, on-the-fly feature-based LRs. By restoring the LR as a comparative measure of similarity and typicality, the proposed framework achieves end-to-end calibration without requiring post-processing. Evaluated on glass fragment LA-ICP-MS data, the method yields a 4.5-fold improvement in the log-likelihood ratio cost (C_llr) over comparable systems, significantly outperforming existing feature-based LR approaches. These results establish an interpretable and highly reliable new paradigm for forensic evidence evaluation.

0 citationsRead paper

From specific-source feature-based to common-source score-based likelihood-ratio systems: ranking the stars

Apr 24, 2026

This study addresses the evaluation and selection of source-level likelihood ratio (LR) systems for forensic evidence-to-reference comparison tasks by proposing an integrated analytical framework that balances performance and practical feasibility. The authors employ strictly proper scoring rules to quantify how effectively each system updates Bayesian prior odds and present the first systematic comparison among specific-source feature-based, common-source anchored, and unanchored score-based LR approaches. Their findings reveal that specific-source feature-based LRs achieve the highest performance but incur substantial experimental costs, whereas common-source feature-based methods offer strong discriminative power with significantly reduced implementation complexity. All LR systems substantially outperform a baseline relying solely on prior odds. This work thus provides both theoretical grounding and practical guidance for selecting LR systems in forensic practice.

0 citationsRead paper