Safeguarding Creativity: The Impacts and Countermeasures of Data Poisoning in Generative AI Systems
Bibliographic record
Abstract
Deep learning is a type of Artificial Intelligence (AI) that has grown quickly and is now widely used in computer software. As the need for deep learning algorithms that can handle more complicated tasks keeps growing very quickly, the improvements in how efficient and productive these algorithms can be utilized for identifying data poisoning is an important step. Data poisoning refers to altering the legitimate data using various techniques and fabricating the expected results. The poisoning will affect the quality of predictions and reliability of the contents generated. The data sensitive applications which rely on classification models for deriving conclusions will be affected by poisoning resulting in unrealistic conclusions and misguided inferences. Continuous occurrence of poorly classified data and unrealistic conclusions will affect the reliability of the system making it less trustworthy. This research looks into long lasting effects of poisoning attacks, which target the training data used in deep learning and assess the way a deep learning model classifies information. Experiments were done to measure the accuracy of different machine learning algorithms in effectively detecting poisoned data. The evaluations were done using KNN, SVM and Logistic Regression models. The analysis concludes that KNN offers better prediction compared to SVM and Logistic Regression. Logistic regression offers less accurate predictions making it unfit for poisoning detection. Proper classification of poisoned data will help research community and users to be aware of legitimacy of data and thereby help them to choose the right data and information for processing, analytics, and deriving conclusions. This will greatly improve the performance and trustworthiness of related models including generative AI systems and makes the learning process un biased which will highly benefit the research community in general.
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 | 0.009 | 0.050 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.002 | 0.006 |
| Scholarly communication | 0.005 | 0.007 |
| Open science | 0.003 | 0.006 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".