Dynamic Split Federated Learning for resource-constrained IoT systems
Bibliographic record
Abstract
Efficient resource utilization in Internet of Things (IoT) systems is challenging due to device limitations. These limitations restrict on-device machine learning (ML) model training, leading to longer processing times and inefficient metadata analysis. Additionally, conventional centralized data collection poses privacy concerns, as raw data has to leave the device to the server for processing. Combining Federated Learning (FL) and Split Learning (SL) offers a promising solution by enabling effective machine learning on resource-constrained devices while preserving user privacy. However, the dynamic nature of IoT resources and device heterogeneity can complicate the application of these solutions, as some IoT devices cannot complete the training task on time. To address these concerns, we have developed a Dynamic Split Federated Learning (DSFL) architecture that dynamically adjusts to the real-time resource availability of individual clients. Combining real-time split-point selection with a Genetic Algorithm (GA) for client selection, tailored to heterogeneous, resource-constrained IoT devices. DSFL ensures optimal utilization of resources and efficient training across heterogeneous IoT devices and servers. Our architecture detects each IoT device’s training capabilities by identifying the number of layers it can train. Moreover, an effective Genetic Algorithm (GA) process strategically selects the clients required to complete the split federated learning round. Cooperatively, the clients and servers train their parts of the model, aggregate them, and then combine the results before moving to the next round. Our proposed architecture enables collaborative model training across devices while preserving data privacy by combining FL’s parallelism with SL’s dynamic modeling. We evaluated our architecture on sensory and image-based datasets, showing improved accuracy and reduced overhead compared to baseline 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 | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 0.002 |
| 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".