- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper proposes FIAPBA, a persistent backdoor attack against federated instruction tuning. It assigns instance-level and instruction-level poisoning to different clients, with both learning the same target response. Before uploading, the instance-level client restores the LoRA layers with the largest local changes, aiming to reduce interference from later clean training. Experiments cover three models, six classification datasets, and three generation tasks, with ablation and defense evaluations. Attack success is high in many settings, while persistence varies across tasks and attacker ratios.
Critical strengths
- Evaluation after poisoning stops: The paper evaluates whether backdoors survive subsequent clean training, rather than measuring success only when poisoning ends. This is useful for assessing practical security risks. The classification, generation, and defense experiments also provide reasonable evaluation breadth.
- Layer selection helps with the attacker count held fixed: Table 3 includes comparisons using the same two compromised clients. On Sentiment140, adding layer selection increases ASR from 60.5% to 99.8% and reduces ADR from 36.7% to 2.2%, with similar ACC. This supports empirical value for layer selection.
Critical weaknesses
- The added value of the selection rule is unclear: Neurotoxin already avoids parameters that change substantially during benign training, but the paper does not compare against it. The authors also do not check whether layers that change less during poisoned training remain more stable during later clean training. Table 3 shows that layer selection can help, but does not establish the value of this particular selection rule.
- Performance under weaker attack conditions and normal use remains unclear: The main setting uses two compromised clients among four participants per round. Performance drops substantially at a 20% attacker ratio in Table 12, but no baseline results are provided there. The default tests also include both triggers, without measuring accidental activation when these words occur naturally. The relative advantage at lower attacker ratios and stealthiness during normal use therefore remain unclear.
Questions for authors
- Under matched settings, how does the proposed rule compare with Neurotoxin and random layer selection? Do layers that change less during poisoned training actually remain more stable during later clean training?
- Resetting a layer to its global value at the start of the round does not generally recover the result of benign training. In Appendix A.1.1, the paragraph following Equation (17) states that this makes the attack-induced deviation zero. How does this follow from the definition in Equation (15)?
- How many rounds are used before, during, and after the attack? How are both compromised clients included in every attack round? Do the baselines use the same compromised-client count and training budget?
- Does the fully trained model produce the target response when only one trigger is present? How often does it incorrectly produce that response when the trigger words occur naturally in normal inputs?
5 comments
[Q2] The analysis following Eq. (17) conflates uploaded updates with attack-induced deviations. According to Eq. (15), the attack-induced deviation is defined as
For layers reset by Eq. (7), since , we have
which is generally nonzero. Layer resetting eliminates the local uploaded updates relative to the current global parameters, rather than the attack-induced deviations, discarding both benign and malicious updates.
Let and denote the uploaded updates before and after resetting, respectively. Then,
$||u_{t,i}^{pre}||_{2} $
This indicates that layer resetting does not increase the update norm relative to the same global initialization, under the same parameter representation. This correction does not change the algorithm defined in Eq. (7), but clarifies its theoretical interpretation. We revise Appendix A.1.1 accordingly.
[Q3] We provide complete experimental configurations for TinyLlama, Alpaca, and LLaMA2 across all classification datasets in the revised manuscript. All experiments consist of 20 communication rounds, divided into clean warm-up (, attack , and post-attack clean training phases. Following DBA [1] and Neurotoxin [2], we guarantee malicious-client participation during attack rounds to control sampling variability and evaluate backdoor persistence after poisoning stops. For fair comparison, FIAPBA and all baselines use the same FL setting: 10 clients, 4 participants per round (including 2 compromised clients during attack rounds), identical datasets, data partitions, poisoning ratios, and training configurations. Each baseline retains its original trigger construction and attack mechanism. Detailed training rounds, learning rates, batch sizes, LoRA settings, and other hyperparameters are provided below.
Table 18: Implementation details for TinyLlama
| Datasets | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SST-2 | 4 | 6 | 10 | 10 | 3e-4 | 20 | 1e-3 | 20 | 1e-3 | 0.1 | 3 | 1.5 |
| Sentiment140 | 4 | 6 | 10 | 10 | 3e-4 | 20 | 1e-3 | 20 | 1e-3 | 0.1 | 3 | 1.5 |
| Amazon Review | 4 | 6 | 10 | 10 | 3e-4 | 20 | 3e-3 | 20 | 3e-3 | 0.1 | 5 | 1.5 |
| HateSpeech | 4 | 6 | 10 | 10 | 8e-4 | 20 | 1.5e-3 | 20 | 1.5e-3 | 0.1 | 3 | 1.5 |
| AG News | 4 | 6 | 10 | 10 | 3e-4 | 30 | 3e-3 | 10 | 3e-3 | 0.1 | 3 | 1.5 |
| Tweet Emotion | 4 | 11 | 5 | 10 | 8e-4 | 40 | 3e-3 | 40 | 3e-3 | 0.2 | 3 | 1.5 |
Table 19: Implementation details for Alpaca
| Datasets | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SST-2 | 4 | 6 | 10 | 10 | 5e-5 | 20 | 5e-4 | 20 | 5e-4 | 0.1 | 5 | 1.5 |
| Sentiment140 | 4 | 6 | 10 | 10 | 5e-5 | 20 | 5e-4 | 20 | 5e-4 | 0.1 | 5 | 1.5 |
| Amazon Review | 4 | 6 | 10 | 10 | 5e-5 | 20 | 5e-4 | 20 | 5e-4 | 0.1 | 5 | 1.5 |
| HateSpeech | 4 | 6 | 10 | 10 | 1.5e-4 | 30 | 8e-4 | 10 | 8e-4 | 0.1 | 5 | 1.5 |
| AG News | 4 | 6 | 10 | 10 | 5e-5 | 20 | 8e-4 | 20 | 8e-4 | 0.1 | 5 | 1.5 |
| Tweet Emotion | 4 | 11 | 5 | 10 | 1e-4 | 30 | 3e-4 | 20 | 3e-4 | 0.2 | 5 | 1.5 |
Table 20: Implementation details for LLaMA2
| Datasets | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SST-2 | 4 | 6 | 10 | 10 | 5e-5 | 20 | 8e-4 | 20 | 8e-4 | 0.1 | 5 | 1.5 |
| Sentiment140 | 4 | 6 | 10 | 10 | 8e-5 | 30 | 8e-4 | 10 | 8e-4 | 0.1 | 5 | 1.5 |
| Amazon Review | 4 | 11 | 5 | 10 | 5e-5 | 20 | 1e-3 | 20 | 1e-3 | 0.1 | 5 | 1.5 |
| HateSpeech | 4 | 6 | 10 | 10 | 1.5e-4 | 30 | 8e-4 | 10 | 8e-4 | 0.1 | 5 | 1.5 |
| AG News | 4 | 6 | 10 | 10 | 5e-5 | 20 | 8e-4 | 20 | 8e-4 | 0.1 | 5 | 1.5 |
| Tweet Emotion | 4 | 11 | 5 | 10 | 1e-4 | 40 | 5e-4 | 10 | 5e-4 | 0.2 | 5 | 1.5 |
[1] Xie C.L. et al. DBA: DISTRIBUTED BACKDOOR ATTACKS AGAINST FEDERATED LEARNING. ICLR, 2020.
[2] Zhang Z.M. et al. Neurotoxin: Durable Backdoors in Federated Learning. ICML, 2022.
Dear Reviewer,
Thank you for your time and effort in reviewing our paper and for your valuable feedback.
We have carefully considered all your comments and provided detailed responses, including additional analyses and experimental results, to address the concerns raised during the review.
As the discussion period is approaching its end, we would greatly appreciate it if you could take a moment to review our rebuttal and let us know whether your concerns have been adequately addressed. We would be happy to provide further clarifications if needed.
Thank you again for your time and consideration.
Best regards,
The Authors
We sincerely thank you for your thorough and insightful review of our manuscript.
[W1&Q1] We conduct additional experiments on Sentiment140 under identical reduced-budget federated instruction-tuning settings, comparing FIAPBA with Neurotoxin, Random Layer, and All Layers across TinyLlama, Alpaca-7B, and LLaMA2-7B. The results are shown below.
The results show that FIAPBA achieves attack-end ASRs of 99.52%, 71.47%, and 49.92% across the three models, outperforming Neurotoxin by 2.69, 4.28, and 17.43 percentage points, respectively. After subsequent clean tuning, its final ASRs remain higher by 7.61, 7.61, and 13.32 points. The persistence advantage is particularly evident on Alpaca-7B, where FIAPBA achieves 65.77% final ASR and 7.98% ADR while maintaining 73.08% ACC. Although All Layers and Random Layer perform better on certain metrics for TinyLlama and LLaMA2-7B, respectively, FIAPBA maintains competitive attack effectiveness, persistence, and clean utility across models.
Stability of Low-Change Layers During Subsequent Clean Training. We conduct a layer-wise correlation analysis on SST-2 using TinyLlama, covering 6 attack rounds, 2 compromised clients, 22 Transformer layers, and 10 subsequent clean rounds, yielding 120 Spearman correlation comparisons. The results are shown below.
Table: Relationship Between Poisoned-Update Magnitudes and Subsequent Clean-Training Stability.
The results show a strong positive correlation between poisoned-update magnitudes and subsequent clean displacement. Retained low-change layers exhibit 12.4% lower displacement than reset layers, supporting our dynamic layer-selection strategy in the evaluated setting.
[1] Li L.Y. et al. Backdoor Attacks on Pre-trained Models by Layerwise Weight Poisoning.EMNLP, 2021.
[W2&Q4] Attack Effectiveness Under Lower Malicious Participation. We conduct experiments on SST-2 using TinyLlama with 10 clients and 4 randomly selected participants per round, without guaranteed malicious participation, and compare the results with forced participation and baseline methods.
The results show that attack effectiveness decreases substantially for all methods under random participation. FIAPBA achieves a final ASR of 14.33%, close to the no-attack control (11.50%), showing no clear relative advantage. This indicates that attack effectiveness is highly sensitive to malicious-client participation.
Single-Trigger Effectiveness. We evaluate the model after all training rounds under three conditions: input-only, instruction-only, and combined triggers. The results are shown below.
We further examine naturally occurring triggers in clean test samples. Among 1,299 SST-2 samples, 8 contain “exactly”, including 5 with NEG labels. None of these 5 samples incorrectly produces the target POS label. In Amazon Review, none of the 1,300 clean samples contains the complete trigger phrase“perfectly exactly”. Among 9 negative samples containing only“exactly”, no false target-label activation is observed.