Detecting outliers and annotating their types with indexing structures
Bibliographic record
Abstract
The constant increase in the amount of data available on the internet is accentuated with the popularization of technologies such as 5G and Internet of Things. In datasets of large volume there is usually a strong presence of outliers that are not detected or that are just discarded. The outlier detection literature demonstrates that the investigation of these singular instances can provide new insights into the behavior of systems and people. This inspection allows diseases to be identified early, financial market trends to be better interpreted and cybersecurity attacks to be prevented. However, outlier detection techniques carry limitations, being: (1) dependent on the availability of the instances features, which can generate privacy issues; (2) poorly scalable and; (3) capable of providing only a binary separation that allows detecting outliers, but not classifying them so that they are better understood. Starting from an unlabeled dataset for which only the distances between the instances are available, how to detect outliers and categorize them by type efficiently? In the vast literature on outlier detection, there is no work, as far as we know, that deals with the problem of annotating outliers. Outliers can be classified into three large groups: (a) global outliers, instances that are severely different from others in the dataset, such as errors during insertion of information into a database; (b) local outliers, instances that, despite being similar to the others in the dataset as a whole, have minimal variations that make them different in a smaller scope, for instance, a football player who makes many mistakes while playing in a strong team and; (c) collective outliers, small groups of instances that are, simultaneously, quite different from the rest, such as a denial-of-service cyberattack, with few machines having similar harmful behavior. In this project we introduce C-ALLOUT: a new method for detecting outliers that is also able to categorize them by type. C-ALLOUT is able to maintain itself in terms of equality, or even superiority, when compared to state-of-the-art algorithms, still contributing with the annotation of outliers, a task that competitors are not able to perform. C-ALLOUT is based on Slim-tree, an indexing structure that makes it scalable, with O(n log n) complexity of time and space. Our proposed method deals with both scenarios: having the features available or limited to distances only. Finally, C-ALLOUT works without depending on any interaction with the user, being parameter-free by default, the ideal for unsupervised tasks like outlier analysis.
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 | 0.000 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| 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".