Android Access Control Recommendation as a Deep Learning Task
Bibliographic record
Abstract
Android enforces access control checks to protect sensitive framework APIs. If not properly protected, framework APIs can open the door for malicious apps to access sensitive resources without having the necessary privileges. Unfortunately, as reported in the existing literature, such access control anomalies are prevalent in Android APIs, notably those introduced by customization parties. Therefore, various solutions have been proposed to detect anomalies, particularly those due to inconsistencies in the enforcement of access checks across the Android framework(s). The solutions can be largely divided into two categories: convergence-based techniques which rely on the convergence of two APIs on similar resources, and probabilistic approaches which incorporate additional hints in the form of manually defined structural and semantic code constructs. In this paper, we are motivated by the promising application of using code constructs, beyond convergence as proposed by the probabilistic approaches, to recommend access control enforcement and detect inconsistencies. \n \nSpecifically, we propose a deep learning-based approach that aims to automatically learn the correspondence between various code constructs and access control requirements. To this end, we fine-tune CodeBert on statically derived features from the Android Open Source Project (AOSP). Our feature engineering process addresses various peculiarities that characterize Android implementations. The resulting fine-tuned model can be queried to recommend access control for vendor-customized APIs. \n \nThe fine-tuned model achieves an accuracy of 93%, a precision of 91%, and a recall of 92% in the AOSP data. Additionally, our evaluation of custom ROMs shows that the model is able to not only rediscover previously reported inconsistencies but also discover new ones.
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 | 0.003 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
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".