Pre-trained Encoders for Global Child Development: Transfer Learning Enables Deployment in Data-Scarce Settings

TL;DR

This paper presents a pre-trained encoder for global child development that, through transfer learning, enables effective deployment in data-scarce settings, significantly outperforming traditional models with minimal training data.

Contribution

The first pre-trained encoder for global child development trained on diverse UNICEF data, demonstrating strong few-shot and zero-shot performance in resource-limited environments.

Findings

Achieves 0.65 AUC with only 50 samples, outperforming gradient boosting.

Reaches 0.73 AUC at 500 samples, showing rapid learning.

Zero-shot deployment yields up to 0.84 AUC in unseen countries.

Abstract

A large number of children experience preventable developmental delays each year, yet the deployment of machine learning in new countries has been stymied by a data bottleneck: reliable models require thousands of samples, while new programs begin with fewer than 100. We introduce the first pre-trained encoder for global child development, trained on 357,709 children across 44 countries using UNICEF survey data. With only 50 training samples, the pre-trained encoder achieves an average AUC of 0.65 (95% CI: 0.56-0.72), outperforming cold-start gradient boosting at 0.61 by 8-12% across regions. At N=500, the encoder achieves an AUC of 0.73. Zero-shot deployment to unseen countries achieves AUCs up to 0.84. We apply a transfer learning bound to explain why pre-training diversity enables few-shot generalization. These results establish that pre-trained encoders can transform the feasibility of ML for SDG 4.2.1 monitoring in resource-constrained settings.