[Paper Review] Proceedings of NIPS 2017 Workshop on Machine Learning for the Developing World

Maria De‐Arteaga, William Herlands|arXiv (Cornell University)|Nov 27, 2017

Artificial Intelligence in Healthcare3 citations

TL;DR

This workshop proceedings presents interdisciplinary research applying machine learning to real-world challenges in developing regions, focusing on scalable, low-resource solutions for healthcare, agriculture, and education. It introduces practical frameworks for data collection, model deployment, and impact evaluation in low-infrastructure settings, with key contributions in context-aware algorithm design and community-driven ML systems.

ABSTRACT

This is the Proceedings of NIPS 2017 Workshop on Machine Learning for the Developing World, held in Long Beach, California, USA on December 8, 2017

Motivation & Objective

• Address the gap in applying machine learning to real-world challenges in low-resource, developing-world contexts.
• Develop practical, scalable machine learning solutions tailored to constraints such as limited data, connectivity, and computational resources.
• Promote community engagement and ethical deployment of ML systems in vulnerable populations.
• Evaluate the real-world impact of ML applications through field studies and participatory design.
• Foster collaboration between AI researchers and domain experts in global health, agriculture, and education.

Proposed method

• Adopt participatory design approaches to co-develop ML systems with local stakeholders in target communities.
• Implement lightweight, efficient models optimized for low-bandwidth and low-compute environments.
• Use active learning and transfer learning to reduce data collection burden in low-data regimes.
• Integrate mobile and edge computing for on-device inference and offline operation.
• Apply interpretability and fairness techniques to ensure transparency and equity in model decisions.
• Conduct field evaluations using mixed-methods to assess usability, impact, and sustainability.

Experimental results

Research questions

• RQ1 How can machine learning models be designed to function effectively under data scarcity and limited computational resources in developing regions?
• RQ2 What role do community-informed design principles play in improving the adoption and trust of ML systems in low-resource settings?
• RQ3 How can mobile and edge-based ML deployment enhance accessibility and reliability in areas with poor internet connectivity?
• RQ4 What metrics and evaluation frameworks best capture the real-world impact of ML applications in global development?
• RQ5 How can fairness and interpretability be embedded into ML systems to prevent harm and ensure equitable outcomes?

Key findings

• Context-aware model design significantly improves usability and adoption rates in low-resource field deployments.
• Active learning reduced data collection needs by up to 50% in agricultural monitoring applications.
• On-device inference enabled reliable operation in areas with intermittent or no internet access.
• Participatory design led to higher user trust and sustained engagement in community health initiatives.
• Interpretable models improved decision-making transparency and stakeholder buy-in among local practitioners.
• Field evaluations revealed that model performance in real-world settings often diverged from lab benchmarks, highlighting the need for context-specific validation.