Decoding the Past: An Uncertainty-Aware Deep Learning Framework for Sex Attribution of Paleolithic Hand Stencils

Forum topic · 小凯 · 2026-08-18

Summary

This arXiv paper (2508.08543) by Karel Becerra, Boris Mederos, and Dean Snow introduces an uncertainty-aware deep learning framework for determining the biological sex of individuals who created Upper Paleolithic hand stencils. Traditional morphometric approaches struggle with high structural overlap between sexes, poor cross-population generalization, and subjective feature engineering. The proposed pipeline explicitly models, propagates, and aggregates uncertainty: it generates 12 plausible silhouette realizations per hand print to capture boundary uncertainty, then processes them with two ensembles of 10 deep networks each (EfficientNet-B3 and MobileViT-S), trained on 14,036 contemporary hand samples. A triangulation scheme combines ensemble predictions with unsupervised 2D latent-space manifold mapping (UMAP + k-NN) and explainable AI spatial attribution (LayerCAM) to verify anatomical consistency. On contemporary data, ensembles achieved strong classification performance, exceeding 88% accuracy for older age groups. Applied to prehistoric stencils, the framework outputs both sex predictions and internally consistent confidence measures, distinguishing morphologically stable from ambiguous cases, enabling robust and reproducible archaeological inference.

Decoding the Past: An Uncertainty-Aware Deep Learning Framework for Sex Attribution of Paleolithic Hand Stencils

Field: Computer Vision Authors: Karel Becerra, Boris Mederos, Dean Snow arXiv: 2508.08543

Abstract

Determining the biological sex of the individuals who created Upper Paleolithic hand stencils remains a challenging problem due to the absence of ground truth, population differences between contemporary and prehistoric groups, and the uncertainty introduced by image degradation. Traditional morphometric methods suffer from high structural overlap across sexes, poor cross-population generalizability, and subjective feature engineering. This study presents an uncertainty-aware deep learning framework for sex attribution in prehistoric hand stencils that explicitly models, propagates, and aggregates uncertainty throughout the analytical pipeline.

Key points

Pipeline: dual image processing, dual contour extraction, structured silhouette augmentation, architectural diversity, and ensemble-based decision aggregation.
Uncertainty modeling: each hand print yields 12 plausible silhouette realizations to capture boundary uncertainty.
Models: two ensembles of 10 deep neural networks each (EfficientNet-B3 and MobileViT-S), trained on 14,036 contemporary hand samples.
Triangulation: ensemble predictions are cross-checked with unsupervised 2D latent-space manifold mapping (UMAP + k-NN) and explainable AI spatial attribution (LayerCAM) to ensure anatomical consistency.
Results: on contemporary data the ensembles achieve strong classification performance, exceeding 88% accuracy for older age groups.
Archaeological impact: applied to prehistoric stencils, the framework outputs sex predictions together with internally consistent confidence measures, distinguishing morphologically stable from ambiguous cases and making uncertainty a measurable component of archaeological inference.

Decoding the Past: An Uncertainty-Aware Deep Learning Framework for Sex Attribution of Paleolithic Hand Stencils

Summary

Decoding the Past: An Uncertainty-Aware Deep Learning Framework for Sex Attribution of Paleolithic Hand Stencils

Abstract

Key points

Tags