Backdoor Attacks on Large Language Model Based Semantic Communication Systems
Bibliographic record
Abstract
We propose an efficient backdoor attack on a task-oriented semantic communication system, designed by employing Large Language Models (LLMs) (e.g., BERT and RoBERTa) and self-attention techniques. To achieve a backdoor model, we assume an adversary creates a clean model then embeds backdoors into it in pre-training phases one and two, respectively. The LLM-based semantic communication system is pre-trained in phase one by using a loss function with the standard cross-entropy loss and a smoothness-inducing adversarial component so that it is robust against Gaussian noise and wireless fading. The salient feature of our design is that the proposed backdoor attack remains effective even if the backdoor system is further fine-tuned for different downstream tasks. This attack effectiveness is achieved thanks to a smart training procedure in the pre-training phase two that maps a poisoned input to an output representation (OR) close to a predefined OR. This predefined OR is highly likely to differ from the true OR of the clean input, leading to incorrect prediction for the considered task. Our experiments show that the proposed attack strategy is efficient and stealthy for a wide range of signal-to-noise ratios (SNR) and different types of triggers and tasks. Moreover, we demonstrate that the proposed attack can withstand the model pruning and fine-tuning defense strategies.
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 | 0.000 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.001 | 0.004 |
| Open science | 0.001 | 0.005 |
| 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".