Adaptive Upgrade of Client Resources for Improving the Quality of Federated Learning Model
Bibliographic record
Abstract
Conventional systems are usually constrained to store data in a centralized location. This restriction has either precluded sensitive data from being shared or put its privacy on the line. Alternatively, federated learning (FL) has emerged as a promising privacy-preserving paradigm for exchanging model parameters instead of private data of Internet of Things (IoT) devices known as clients. FL trains a global model by communicating local models generated by selected clients throughout many communication rounds until ensuring high learning performance. In these settings, the FL performance highly depends on selecting the best available clients. This process is strongly related to the quality of their models and their training data. Such selection-based schemes have not been explored yet, particularly regarding participating clients having high-quality data yet with limited resources. To address these challenges, we propose in this article FedAUR, a novel approach for an adaptive upgrade of clients resources in FL. We first introduce a method to measure how a locally generated model affects and improves the global model if selected for aggregation without revealing raw data. Next, based on the significance of each client parameters and the resources of their devices, we design a selection scheme that manages and distributes available resources on the server among the appropriate subset of clients. This client selection and resource allocation problem is thus formulated as an optimization problem, where the purpose is to discover and train in each round the maximum number of samples with the highest quality in order to target the desired performance. Moreover, we present a Kubernetes-based prototype that we implemented to evaluate the performance of the proposed approach.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.007 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.015 | 0.018 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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; both teacher heads agree on what is shown here.
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".