Seeding Deception: Investigating the use of GANs in Minimizing Backdoor Poisoning Attack Ratios
Bibliographic record
Abstract
As machine learning becomes widely adopted in industry for mission-critical systems, significant research is currently dedicated to the concern of cybersecurity attacks on open-source data. As we see an increasing dependency on open-source databases in model training for institutional use, there is a similarly increasing number of ways for attackers to exploit the Internet - the training ground for such machine learning models. Our investigation centers around using a generative adversarial network (GAN) to minimize the number of poisoned data elements in order to trigger a model into generating inaccurate results. In this paper, we focus on one type of adversarial attack - the backdoor attack, where the attacker provides poisoned data to the victim to train the model with, and then activates the attack by showing a specific small trigger pattern at test time (e.g., a small patch of pixels on an image). We aim to use the GAN in order to optimize the trigger pattern (i.e., pixel mask) added to corrupted data samples and minimize the poisoning ratio for a binary image classification convolutional neural network (CNN). The perspective of this research study focuses on the implications of targeted cybersecurity attacks on open-source datasets. It is therefore of utmost importance that researchers understand the mechanics of various machine learning attacks, and push the envelope on state-of-the-art attacks, such that researchers and engineers can proactively create defenses. This is the principal motivation behind white-hat hacking and thus behind this paper.
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 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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.002 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, 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".