Adaptive Weighted Federated Domain Adaptation Methods for Nonintrusive Load Monitoring
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The implementation of scalable and privacy-preserving models in nonintrusive load monitoring (NILM) requires the usage of distributed pipelines through methods like federated learning. However, the differences between building locations, appliances, and consumption patterns create domain discrepancies that hinder performance. To address this issue, we need to apply domain adaptation while respecting the privacy constraint of federated learning. In this work, we propose adopting existing methods to act as discrepancy measures in the context of federated domain adaptation in NILM using a deep sparse coding model with novel modifications. These methods find the discrepancy between the local data and the target data and include: 1) a modified version of Federated domain adaptation (FedDA), which is based on the local variance and Euclidean distance; 2) FedRBF, which is based on the measure of similarity using the radial basis function kernel; 3) FedMMD, which is based on maximum mean discrepancy; and 4) FedkNN, which is based on the k-nearest neighbors algorithm. Moreover, we propose a novel framework that adaptively combines all methods as a second intermediate layer by learning their contribution weights to the central server. This deep adaptive weighting framework aims to optimize the selection of the output weights of each method in a federated manner. We explore several techniques of weight computation that are based on the raw data or on learned representations that further preserve user consumption privacy. We find that every method performs better than regular federated averaging (FedAvg) by up to 5% in terms of disaggregation accuracy, while an additional increase for the adaptive weighting framework results. The suggested methods also perform better than training only on the target domain. We also find that basing the weights on locally learned representations yields similar results to using raw data.
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.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 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 it