Mitigating Membership Inference Attacks in Machine Learning as a Service
Bibliographic record
Abstract
The increasing use of Machine Learning as a Service (MLaaS) has raised privacy and security issues due to membership inference attacks. These attacks can extract sensitive information such as the identification of an individual's participation in a training dataset, by exploiting a binary classifier with limited access. The attacks exploit weaknesses in the decision boundaries of the model, and can lead to the disclosure of private information. However, the current defenses against such attacks, such as those based on differential privacy or regularization, have significant limitations. Therefore, further research is needed to develop effective defenses that maintain the utility of machine learning models while providing formal guarantees, even in the presence of strategic adversaries. In this paper, we focus on mitigating the risks of black-box inference attacks against machine learning models as a service. We propose a defense mechanism that brings the attacker's inference classifier into a zone of uncertainty, rendering it unable to classify a data point as a member or non-member. This mechanism takes into account the attacker's behavior by modeling the interaction between defense and attacker as a game, considering potential gains in confidentiality and costs. Our experiments on two datasets demonstrate the effectiveness of our approach in mitigating membership inference attacks. Furthermore, our defense mechanism outperforms existing defenses by offering superior privacy-utility-performance tradeoffs.
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 | 0.016 | 0.043 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.002 | 0.003 |
| Scholarly communication | 0.004 | 0.011 |
| Open science | 0.004 | 0.009 |
| Research integrity | 0.004 | 0.007 |
| 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".