A Comparative Study of Federated Learning and Synthetic Data for Privacy-Aware Machine Learning
Bibliographic record
Abstract
Healthcare institutions face a critical challenge in training and deploying machine learning applications due to data scarcity compounded by stringent privacy regulations. In this case study involving breast cancer identification, we evaluated four experimental scenarios under conditions of limited data availability and strict privacy requirements. Specifically, we compared: (i) federated learning with distributed real data, (ii) federated learning with synthetic data, (iii) centralized learning on aggregated synthetic datasets generated locally, and (iv) multi-step synthetic data generation. Our results indicate that when local datasets are too small to be useful independently, federated learning with real data achieves the highest performance, outperforming federated learning with synthetic data. In contrast, models developed on aggregated synthetic datasets or via centralized generation of synthetic data based on local synthetic samples yielded suboptimal results. Although federated learning with real data appeared to be the best-performing strategy, it still fell behind centralized learning with pooled real data. This result demonstrates that federated learning is preferable to synthetic data approaches in low dataset scenarios. Additionally, the modest performance gap compared to the centralized real-data benchmark underscores the importance of further research into improved federated methods.
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 | 0.010 | 0.023 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".