LENS: Low-Resource ML for Social Impact

In “Machine Learning with Limited Resources for Positive Social Impact” we investigate machine learning methods designed to operate effectively under limited resource constraints, with a direct focus on generating positive social impact. Resource limitations are addressed across multiple dimensions — including scarce labeled data, low-compute environments, and underrepresented languages — to enable the deployment of intelligent systems in contexts that are typically overlooked by mainstream AI research.

The project draws on techniques such as transfer learning, few-shot and zero-shot learning, data augmentation, and semi-supervised learning, applying them to socially relevant problems including hate speech detection, health misinformation, accessibility, and vulnerable population support — with a strong emphasis on Brazilian Portuguese and the regional realities of Rio de Janeiro.

Objectives

• Design and evaluate ML algorithms that achieve strong performance with minimal labeled data
• Develop resources and benchmarks for low-resource NLP in Brazilian Portuguese
• Apply limited-resource methods to high-impact tasks in health, safety, and social inclusion
• Contribute open-source tools and datasets to the broader research community

Funding & Support

This project is financed by FAPERJ (Carlos Chagas Filho Foundation for Research Support of the State of Rio de Janeiro) under the Young Scientist from Rio de Janeiro program, covering the period from March 2023 to October 2026.