Design and Development of AD-CGAN: Conditional Generative Adversarial Networks for Anomaly Detection
Bibliographic record
Abstract
Whether in the realm of software or hardware, datasets representing the state of systems are mostly imbalanced. This imbalance is because these systems' reliability requirements make the occurrence of an anomaly a rare phenomenon. Hence, most datasets on anomaly detection have a relatively small percentage that captures the anomaly. Recently, generative adversarial networks (GAN) have shown promising results in image generation tasks. Therefore, in this research work, we build on conditional GANs (CGAN) to generate plausible distributions of a given profile to solve the challenge of data imbalance in anomaly detection tasks and present a novel framework for anomaly detection. Firstly, we learn the pattern of the minority class data samples using a single class CGAN. Secondly, we use the knowledge base of the single class CGAN to generate samples that augment the minority class samples so that a binary class CGAN can train on the typical and malicious profiles with a balanced dataset. This approach inherently eliminates the bias imposed on algorithms from the dataset and results in a robust framework with improved generalization. Thirdly, the binary class CGAN generates a knowledge base that we use to construct the cluster-based anomaly detector. During testing, we do not use the single class CGAN, thereby providing us with a lean and efficient algorithm for anomaly detection that can do anomaly detection on semi-supervised and non-parametric multivariate data. We test the framework on logs and image-based anomaly detection datasets with class imbalance. We compare the performance of AD-CGAN with GAN-derived and non-GAN-derived state of the art algorithms on benchmark datasets. AD-CGAN outperforms most of the algorithms in the standard metrics of Precision, Recall, and F-1 Score. Where AD-CGAN does not perform better in the parameters used, it has the advantage of being lightweight. Therefore, it can be deployed for both online and offline anomaly detection tasks since it does not use an input sample inversion strategy.
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 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.003 | 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".