Federated Learning for cross-jurisdictional analyses: A case study.
Bibliographic record
Abstract
The objective of this project is to implement a harmonized artificial intelligence (AI)-based de-identification of free-text medical data across multiple Canadian jurisdictions. This federated learning approach will allow these jurisdictions to leverage each other’s data and resources while no individual-level data leaves the jurisdiction. Federated Learning enables health data centers in different jurisdictions to collaborate in training machine learning models without sharing individual-level data. This approach will significantly reduce privacy and cybersecurity risks and barriers that are involved in sharing and moving data across different jurisdictions. In a federated learning environment, machine learning models are trained on multiple data sources available in local data centers; local data are not shared to a central computing/analysis environment. Instead, parameters (such as model weights) are shared between these local data centers to generate a global model that will be shared and used by all participating data centers. In this case study, four health research data centers in different Canadian provinces will take part in deployment of an AI-based application for de-identification of free-text data. The data centers are members of Health Data Research Network (HDRN) Canada. The deployment will include: harmonized annotation and labeling of local data, local training of entity recognition algorithms, integrating model weights from each data centers to create a global model development of license agreements between the participating data centers to allow sharing model weights This is an ongoing project. The talk will demonstrate learning experiences, advantages, and challenges in a federated learning environment and explore the feasibility of transporting this approach to other multi-jurisdiction research networks.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.019 | 0.031 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.002 | 0.004 |
| Science and technology studies | 0.005 | 0.002 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.003 | 0.004 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".