Enhancing the Performance of Model Pruning in Over-the-Air Federated Learning with Non-IID Data
Bibliographic record
Abstract
This paper focuses on over-the-air federated learning (OTA-FL) for edge devices that have non-independent and identically distributed (non-IID) datasets. Federated averaging (FedAvg), the vanilla FL algorithm does not perform well with non-IID data, leading to reduced global model performance. Pruning, a deep neural network model compression technique can reduce model size to be compatible with low-complexity devices without a significant impact on accuracy. However, inclusion of pruning and channel impairments can significantly degrade the performance of OTA-FL networks in a non-IID setup. We therefore first investigate and quantify this pruning-accuracy trade-off when edge devices utilize an OTA-FL setup - and find significant degradation for pruning rates over 50%. We then investigate two approaches to mitigate this degradation. The first is to utilize FedProx, an effective federated aggregation algorithm designed to improve performance on edge devices with non-IID data. The second approach is to iteratively re-train the model at the parameter server (PS) using a limited dataset representative of the task being learned across the edge devices. This approach is suited for cases where it is possible to access such data, e.g. keyboard prediction, several computer vision and natural language processing tasks. We conduct thorough simulations to assess the performance and convergence behavior of both approaches, for different pruning levels and datasets. Results show that considerable gains can be achieved with these approaches, particularly when pruning is not too aggressive and for less complex classification tasks. Further research is however needed to develop algorithms that are more robust to high pruning, non-IID data distributions, and OTA aggregation at low signal to noise ratios (SNRs).
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 | 0.003 | 0.012 |
| 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 | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.002 |
| 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".