Learning probabilistic dependencies among events for proactive security auditing in clouds
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Security compliance auditing is a viable solution to ensure the accountability and transparency of a cloud provider to its tenants. However, the sheer size of a cloud, coupled with the high operational complexity implied by the multi-tenancy and self-service nature, can easily render existing runtime auditing techniques too expensive and non-scalable. To this end, a proactive approach, which prepares for the auditing ahead of critical events, is a promising solution to reduce the response time to a practical level. However, a key limitation of such approaches is their reliance on manual efforts to extract the dependency relationships among events, which greatly restricts their practicality. What makes things worse is the fact that, as the most important input to security auditing, the logs and configuration databases of a real world cloud platform can be unstructured and not ready to be used for efficient security auditing. In this paper, we first propose a log processing technique, which prepares raw cloud logs for different analysis purposes, and then design a learning-based proactive security auditing system, namely, [Formula: see text]. To this end, we conduct case studies on current log formats in different real-world OpenStack (a popular cloud platform) deployments, and identify major challenges in log processing. Later, we design a stand-alone log processor for clouds, which may potentially be used for various log analyses. Consequently, we leverage the log processor outputs to extract probabilistic dependencies from runtime events for the dependency models. Finally, through these dependency models, we proactively prepare for security critical events and prevent security violations resulting from those critical events. Furthermore, we integrate [Formula: see text] to OpenStack and perform extensive experiments in both simulated and real cloud environments that show a practical response time (e.g., 6 ms to audit a cloud of 100,000 VMs) and a significant improvement (e.g., about 50% faster) over existing proactive approaches. In addition, we successfully and efficiently apply our log processor outputs to other learning techniques (e.g., executing sequence pattern mining algorithms within 18 ms for 50,000 events).
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.
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 | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| 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 it