AI for global health

CS-467 / 8 crédits

Langue: Anglais

Withdrawal: It is not allowed to withdraw from this subject after the registration deadline.

Summary

This course teaches students to design, build and rigorously evaluate trustworthy, representative and contextually appropriate AI systems for high-stakes decision-making in global health and humanitarian response.

Content

Paired lectures. The course follows a paired-lecture, call-and-response format. Each week there is a:

Domain lecture: defining a consequential global-health or humanitarian problem, including its context, stakeholders, constraints and uncertainties.
Technical lecture: developing a principled engineering response, examining appropriate methods, implementation choices, evidence requirements and potential failure modes.

Real world group project. A weekly project session allows students to work in groups of 4-5 to apply course concept to real world humanitarian or global health project.

Three red threads

Volatile Contexts. Understanding the environments in which AI must operate and why technologies may fail in humanitarian and clinical settings.
High-Stakes Decisions. Understanding how professionals reason under uncertainty and how AI can safely support human expertise.
Trustworthy Evidence. Determining whether AI systems are safe, effective, contextually appropriate and ready for real-world use at scale.

Domain lectures may address:

Humanitarian systems and humanitarian response
Global health and community medicine
Health systems and resource-constrained settings
Epidemics and outbreak response
Clinical reasoning, diagnosis and triage
Evidence-based medicine and clinical workflows
Human decision-making under uncertainty
Clinical trials and evidence generation
Implementation science
Regulation, governance and responsible innovation

Technical lectures may address:

Participatory and contextually appropriate system design
Data quality, representativeness and reproducibility
Machine-learning infrastructure and software engineering
Privacy-preserving methods and federated learning
Edge computing and constrained deployment
Data analysis and predictive modelling
Bayesian reasoning and causal inference
Retrieval-augmented generation
Clinical and humanitarian decision-support systems
Human-AI collaboration and calibrated uncertainty
Validation, benchmarking and subgroup evaluation
Safety, deployment, monitoring and MLOps
Technical communication and development for scale

240 hours total including

Keywords

Artificial intelligence; machine learning; global health; humanitarian response; clinical decision support; medicine; AI4Good; trustworthy AI; representative AI; responsible AI; uncertainty; causal inference; fairness; ethics; implementation science; privacy; federated learning; edge computing; MLOps.

Learning Prerequisites

Required courses

Introduction to machine learning (CS-233) or Biological data science II: machine learning (BIO-322) or Applied data analysis (CS-401) or equivalent

Recommended courses

- Machine learning (CS-433)
Applied data analysis (CS-401)
Modern natural language processing (CS-552)
Digital epidemiology (BIO-512)
Understanding statistics and experimental design (BIO-449)
Ethics for life sciences engineers (BIO-508)

It is helpful to have completed a semester project prior to this course to learn the art of project management.

Important concepts to start the course

Basic probability and statistics
Basic machine learning concepts and model evaluation
Basic programming skills, preferably in Python
Basic software engineering practices for collaborative projects
Interest in health, humanitarian response and responsible technology

Learning Outcomes

By the end of the course, the student must be able to:

Formulate AI problems under explicit global-health, humanitarian and operational constraints.
Design AI systems grounded in real-world deployment and validation requirements.
Integrate ethical, regulatory and governance requirements into technical design decisions.
Anticipate uncertainty, distribution shifts, failure modes and downstream harms.
Construct rigorous evaluation strategies using validation, benchmarking, calibration and subgroup analysis.
Critique claims about the performance, safety, trustworthiness and impact of AI systems.
Distinguish model performance from usability, safety and system-level impact.
Present an AI system's intended use, supporting evidence, limitations and remaining uncertainties.

Teaching methods

Interactive lectures combining domain and engineering perspectives
Guest lectures from global-health and humanitarian practitioners
A semester-long external group project involving teams of four to five students
Structured interdisciplinary paper analysis and presentation
Design reviews and stakeholder-feedback sessions
Project presentations, discussions, and live questions
Certification-based ethics and research-compliance learning

Expected student activities

Attend lectures and participate actively in discussions and critique sessions.
Work continuously on a semester-long interdisciplinary group project.
Design and revise an AI system under evolving technical and contextual constraints.
Analyze and present interdisciplinary scientific papers.
Present intermediate and final project results.
Complete the required ethics and research-compliance certifications.
Maintain reproducible code and appropriate technical documentation.
Submit a final project report describing the system, evidence, limitations, and implications

Assessment methods

Continuous assessment

In-class written quizzes on lecture material
Initial submission: project concept
Midterm checkpoint: proof of concept of the project (reproducible code)
Midterm oral presentation with live questions
Individual final project report and code (git commit audit), including the final presentation