Interactive Machine Learning in Cybersecurity: Using Human Expertise More Effectively
Bibliographic record
Abstract
Cybersecurity is increasingly important in a world where malicious actors seek to profit from various forms of data exfiltration. In this dissertation I examine the potential for improving the detection of anomalies related to potential data exfiltration in a large financial service company through better use of human expertise. The overall approach that guided the research reported below is interactive Machine Learning (iML) where humans work together with Machine Learning to solve prediction and classification problems. After reviewing work on data exfiltration and ML, I carried out a series of three case studies with the overall goal of using iML to improve defence against data exfiltration. I first demonstrated that the best known synthetic data set did not provide credible results that would generalize to real world settings. I then did two further case studies on real world data (from a financial services company). In the first of those further studies I looked at how the organization of human labelers affects outcomes with active learning (AL), comparing model training performance between individuals, groups, and situations where pairs of labelers exchanged places halfway through the training process so that different people trained the model on the second half of the training process. I found that a single individual performed best and the role of expertise was further reinforced in my final case study where a group of skilled (vs. unskilled) labelers was shown to produce higher model performance, as well as producing confidence ratings in their labels that were better calibrated with resulting model accuracy. I also showed that an information gain maximizing approach was a viable method of AL in this application area, with better F1 scores and labeler confidence than a more traditional model uncertainty approach. One limitation of this research was that due to privacy concerns I was using redacted email data that did not contain the email body.
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 | 0.011 | 0.029 |
| 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 | 0.003 |
| Scholarly communication | 0.004 | 0.007 |
| Open science | 0.002 | 0.005 |
| Research integrity | 0.002 | 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".