Poisoning Attack Mitigation for Privacy-Preserving Federated Learning-Based Energy Theft Detection
Bibliographic record
Abstract
In federated learning (FL) based electricity theft detection, detection nodes (DNs) locally train deep learning models on consumers' data and share only the local model parameters with an aggregation server (AS) to generate a global model shared by all nodes for better detection accuracy. However, several privacy concerns should be addressed including membership and inference attacks. To mitigate these attacks, several privacy-preserving aggregation schemes have been introduced. Nevertheless, existing FL detectors often overlook the threat of poisoning attacks, in which certain DNs hold maliciously labeled, i.e., poisoned, data during the training. This manipulated data can subsequently be exploited to introduce backdoors into the global model after its deployment. This paper introduces a novel approach that enhances privacy and resilience against poisoning attacks in FL-based electricity theft detection within smart grids. Our approach enables encrypting local parameters before sending them to the AS, thus safeguarding consumers' privacy. Additionally, it utilizes a cosine similarity test over encrypted data to detect and mitigate poisoning attacks by filtering out malicious local gradients from being considered in the global model computation. Through extensive evaluations, we demonstrate the effectiveness of our FL-based detector in substantially reducing the poisoning attack success rate even when 50% of DNs train their local models with malicious targeted power consumption data, all while preserving consumers' privacy.
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 | 0.002 | 0.007 |
| 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 | 0.002 |
| 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".