Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

JAM: A Comprehensive Model for Age Estimation, Verification, and Comparability

François David, Alexey A. Novikov, Ruslan Parkhomenko, Artem Voronin, Alix Melchy

TL;DR

JAM addresses age estimation, verification, and comparability by delivering distribution-based predictions that output a mean age $y_{\mu}$ and uncertainty $y_{\sigma}$, enabling probabilistic confidence intervals. A differentiable loss $L_{\text{jam}} = \alpha L_{\text{reg}} + \beta L_{\text{std}} + \delta L_{\text{dist}}$ with an age-decay factor $AD_i$ and max-age normalization $M$ drives joint optimization of the mean and variance. The framework demonstrates improved accuracy and reduced false positives across real-world (JPD) and public (ONOT) data, with strong NIST FATE performance and clear benefits for age verification and comparability tasks. The work emphasizes practical deployment considerations, including privacy, consent, bias mitigation, and responsible usage guidelines, aiming for robust, consent-driven age-assurance systems.

Abstract

This paper introduces a comprehensive model for age estimation, verification, and comparability, offering a comprehensive solution for a wide range of applications. It employs advanced learning techniques to understand age distribution and uses confidence scores to create probabilistic age ranges, enhancing its ability to handle ambiguous cases. The model has been tested on both proprietary and public datasets and compared against one of the top-performing models in the field. Additionally, it has recently been evaluated by NIST as part of the FATE challenge, achieving top places in many categories.

JAM: A Comprehensive Model for Age Estimation, Verification, and Comparability

TL;DR

JAM addresses age estimation, verification, and comparability by delivering distribution-based predictions that output a mean age and uncertainty , enabling probabilistic confidence intervals. A differentiable loss with an age-decay factor and max-age normalization drives joint optimization of the mean and variance. The framework demonstrates improved accuracy and reduced false positives across real-world (JPD) and public (ONOT) data, with strong NIST FATE performance and clear benefits for age verification and comparability tasks. The work emphasizes practical deployment considerations, including privacy, consent, bias mitigation, and responsible usage guidelines, aiming for robust, consent-driven age-assurance systems.

Abstract

This paper introduces a comprehensive model for age estimation, verification, and comparability, offering a comprehensive solution for a wide range of applications. It employs advanced learning techniques to understand age distribution and uses confidence scores to create probabilistic age ranges, enhancing its ability to handle ambiguous cases. The model has been tested on both proprietary and public datasets and compared against one of the top-performing models in the field. Additionally, it has recently been evaluated by NIST as part of the FATE challenge, achieving top places in many categories.
Paper Structure (16 sections, 7 equations, 3 figures, 3 tables)

This paper contains 16 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHODOLOGY
  4. Breakdown of the Loss Function Components
  5. Testing & Inference
  6. RESULTS AND DISCUSSION
  7. Setup
  8. Datasets
  9. Age Estimation Results
  10. Age Verification Results
  11. Age Comparability Results
  12. NIST testing
  13. CONCLUSION
  14. Risk
  15. Mitigation
  16. ...and 1 more sections

Figures (3)

  • Figure 1: A visual of the model's age probability functions and thresholds shows significant shifts with varying strictness. The pink area marks the accepted age range for a hypothetical age comparability task.
  • Figure 2: Comparison of MAE scores between our model and the MiVOLO model across various regions. We used a regional breakdown similar to that in the NIST evaluations but included more regions
  • Figure 3: Visual representation of the age comparability ranges obtained with and without taking into consideration the confidence. Tuned for 0.5% FPR on the JPD Dataset.