Efficient anomaly detection using diffusion
Bibliographic record
Abstract
Anomaly detection is the identification of unusual data by assuming that the normal data follows a specific pattern.It plays a crucial role in various domains such as medical monitoring, cybersecurity, and fraud detection.Identifying anomalies becomes challenging when their distribution is unknown, leading to the absence of labelled data during training.Diffusion models are attractive candidates since they are great at learning the data distribution.Once learned, we can leverage the distribution of the normal data to identify anomalies.Previously, the reconstruction error has been used for this purpose, but it is computationally expensive due to the need for multiple predictions.Instead, we propose a method that directly learns the diffusion time, estimating how far an instance is from the data manifold, thus improving efficiency and speed.We empirically demonstrate competitive performance against multiple baselines on a comprehensive anomaly detection benchmark.iv Résumé La détection d'anomalies consiste à identifier des données inhabituelles en supposant que les données normales suivent un modèle de distribution spécifique.Elle joue un rôle crucial dans divers domaines tels que la surveillance médicale, la cybersécurité et la détection de fraudes.L'identification des anomalies devient difficile lorsque leur distribution est inconnue, conduisant à l'absence de données étiquetées lors de l'entraînement.Les modèles de diffusion sont des candidats attrayants puisqu'ils sont excellents pour apprendre la distribution des données.Une fois apprise, on peut exploiter la distribution des données normales pour identifier les anomalies.Auparavant, l'erreur de reconstruction était utilisée à cette fin, mais elle est coûteuse en termes de calcul en raison de la nécessité de faire plusieurs prédictions.Plutôt, nous proposons une méthode qui apprend directement le temps de diffusion, estimant la distance entre une instance et la variété des données, améliorant ainsi l'efficacité et la vitesse.Nous démontrons empiriquement des performances concurrentes à plusieurs algorithmes de référence sur un banc d'essaie complet de détection d'anomalies.
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 | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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".