Your role: Federated Model Developer / Data Scientist

Introduction

The Federated Model Developer or Federated Data Scientist is responsible for designing, training, evaluating, and refining machine learning models in distributed, privacy-preserving environments. Unlike traditional data scientists, they work without direct access to raw data — instead orchestrating model development across multiple, siloed datasets hosted by data partners.

This role requires a combination of ML expertise, creativity, and adaptability. Developers must navigate technical limitations (non-IID data, communication constraints), ensure convergence and performance, and work closely with clinicians, infrastructure teams, and governance leads to ensure models are interpretable, ethical, and compliant.

Key Responsibilities

• Design model architectures suitable for FL use cases (e.g. FedAvg, FedProx, FedBN)
• Select or implement federated learning algorithms that handle heterogeneous data
• Coordinate training across sites using orchestration platforms (e.g. Flower, Substra)
• Tune hyperparameters, monitor convergence, and manage model versioning
• Apply privacy-preserving techniques (e.g. differential privacy, secure aggregation)
• Validate and benchmark models, both locally and globally
• Document training setups, evaluation procedures, and assumptions
• Collaborate with clinical researchers to interpret outputs and assess impact

Common Challenges

• Limited observability into local data (no access to raw datasets)
• Non-IID distributions leading to biased or unstable models
• Communication constraints and system failures across nodes
• Difficulty debugging training without full visibility
• Aligning ML goals with clinical or institutional requirements
• Explaining model behavior to non-technical stakeholders
• Ensuring reproducibility and ethical use of models

Recommended Tools & Resources

FL Frameworks

• Flower - Federated learning framework
• Fed-BioMed - Biomedical federated learning
• Substra - Enterprise federated learning platform
• FedML - Research and production federated learning

Privacy-Preserving Tools

• PySyft - Privacy-preserving machine learning
• TensorFlow Federated - Federated learning for TensorFlow
• TenSEAL - Homomorphic encryption for TensorFlow

Evaluation & Explainability

• SHAP, LIME - Model interpretability
• Custom dashboards for federated validation and metrics aggregation

Benchmark Datasets & Challenges

• Federated Tumor Segmentation (FeTS)
• PATE, LEAF, and other synthetic or split datasets for prototyping

Relevant FLKit Sections

• Enable Infrastructure: deployment and orchestration
• Enhance & Wrangle Data: input schema and harmonisation
• Analyse Shared Data: training, evaluation, reporting
• Plan & Govern: privacy risks, model sharing policies

Training & Further Reading

• OpenMined Courses on Privacy-Preserving ML
• FL Benchmark Papers & Challenges
• Nature Communications: The future of Federated Learning in Healthcare

Solution

• European Data Protection Supervisor’s “Preliminary opinion on Data Protection and Scientific Research”
• BBMRI-ERIC ELSI Knowledge Base contains governance templates and guidance for federated learning projects.
• Data Stewardship Wizard (DSW) can help establish governance frameworks for federated learning projects.
• FAIR Cookbook provides step-by-step recipes for data governance tasks.
• TeSS Training Portal offers training materials on data governance and management.

Related pages

More information

Links to FAIR Cookbook

Links to DSW

Training

Contributors