On the Robustness of the Backdoor-based Watermarking in Deep Neural\n Networks
Bibliographic record
Abstract
Obtaining the state of the art performance of deep learning models imposes a\nhigh cost to model generators, due to the tedious data preparation and the\nsubstantial processing requirements. To protect the model from unauthorized\nre-distribution, watermarking approaches have been introduced in the past\ncouple of years. We investigate the robustness and reliability of\nstate-of-the-art deep neural network watermarking schemes. We focus on\nbackdoor-based watermarking and propose two -- a black-box and a white-box --\nattacks that remove the watermark. Our black-box attack steals the model and\nremoves the watermark with minimum requirements; it just relies on public\nunlabeled data and a black-box access to the classification label. It does not\nneed classification confidences or access to the model's sensitive information\nsuch as the training data set, the trigger set or the model parameters. The\nwhite-box attack, proposes an efficient watermark removal when the parameters\nof the marked model are available; our white-box attack does not require access\nto the labeled data or the trigger set and improves the runtime of the\nblack-box attack up to seventeen times. We as well prove the security\ninadequacy of the backdoor-based watermarking in keeping the watermark\nundetectable by proposing an attack that detects whether a model contains a\nwatermark. Our attacks show that a recipient of a marked model can remove a\nbackdoor-based watermark with significantly less effort than training a new\nmodel and some other techniques are needed to protect against re-distribution\nby a motivated attacker.\n
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 | 0.019 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| 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 | 0.003 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 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".