A large body of work shows that machine learning (ML) models can leak sensitive or confidential information about their training data. Recently, leakage due to distribution inference (or property inference) attacks is gaining attention. In this attack, the goal of an adversary is to infer distributional information about the training data. So far, research on distribution inference has focused on demonstrating successful attacks, with little attention given to identifying the potential causes of the leakage and to proposing mitigations. To bridge this gap, as our main contribution, we theoretically and empirically analyze the sources of information leakage that allows an adversary to perpetrate distribution inference attacks. We identify three sources of leakage: (1) memorizing specific information about the $\mathbb{E}[Y|X]$ (expected label given the feature values) of interest to the adversary, (2) wrong inductive bias of the model, and (3) finiteness of the training data. Next, based on our analysis, we propose principled mitigation techniques against distribution inference attacks. Specifically, we demonstrate that causal learning techniques are more resilient to a particular type of distribution inference risk termed distributional membership inference than associative learning methods. And lastly, we present a formalization of distribution inference that allows for reasoning about more general adversaries than was previously possible.
Algorithms such as Differentially Private SGD enable training machine learning models with formal privacy guarantees. However, there is a discrepancy between the protection that such algorithms guarantee in theory and the protection they afford in practice. An emerging strand of work empirically estimates the protection afforded by differentially private training as a confidence interval for the privacy budget $\varepsilon$ spent on training a model. Existing approaches derive confidence intervals for $\varepsilon$ from confidence intervals for the false positive and false negative rates of membership inference attacks. Unfortunately, obtaining narrow high-confidence intervals for $\epsilon$ using this method requires an impractically large sample size and training as many models as samples. We propose a novel Bayesian method that greatly reduces sample size, and adapt and validate a heuristic to draw more than one sample per trained model. Our Bayesian method exploits the hypothesis testing interpretation of differential privacy to obtain a posterior for $\varepsilon$ (not just a confidence interval) from the joint posterior of the false positive and false negative rates of membership inference attacks. For the same sample size and confidence, we derive confidence intervals for $\varepsilon$ around 40% narrower than prior work. The heuristic, which we adapt from label-only DP, can be used to further reduce the number of trained models needed to get enough samples by up to 2 orders of magnitude.
With the goal of generalizing to out-of-distribution (OOD) data, recent domain generalization methods aim to learn "stable" feature representations whose effect on the output remains invariant across domains. Given the theoretical connection between generalization and privacy, we ask whether better OOD generalization leads to better privacy for machine learning models, where privacy is measured through robustness to membership inference (MI) attacks. In general, we find that the relationship does not hold. Through extensive evaluation on a synthetic dataset and image datasets like MNIST, Fashion-MNIST, and Chest X-rays, we show that a lower OOD generalization gap does not imply better robustness to MI attacks. Instead, privacy benefits are based on the extent to which a model captures the stable features. A model that captures stable features is more robust to MI attacks than models that exhibit better OOD generalization but do not learn stable features. Further, for the same provable differential privacy guarantees, a model that learns stable features provides higher utility as compared to others. Our results offer the first extensive empirical study connecting stable features and privacy, and also have a takeaway for the domain generalization community; MI attack can be used as a complementary metric to measure model quality.
Making evidence based decisions requires data. However for real-world applications, the privacy of data is critical. Using synthetic data which reflects certain statistical properties of the original data preserves the privacy of the original data. To this end, prior works utilize differentially private data release mechanisms to provide formal privacy guarantees. However, such mechanisms have unacceptable privacy vs. utility trade-offs. We propose incorporating causal information into the training process to favorably modify the aforementioned trade-off. We theoretically prove that generative models trained with additional causal knowledge provide stronger differential privacy guarantees. Empirically, we evaluate our solution comparing different models based on variational auto-encoders (VAEs), and show that causal information improves resilience to membership inference, with improvements in downstream utility.
ML-as-a-service is gaining popularity where a cloud server hosts a trained model and offers prediction (inference) service to users. In this setting, our objective is to protect the confidentiality of both the users' input queries as well as the model parameters at the server, with modest computation and communication overhead. Prior solutions primarily propose fine-tuning cryptographic methods to make them efficient for known fixed model architectures. The drawback with this line of approach is that the model itself is never designed to operate with existing efficient cryptographic computations. We observe that the network architecture, internal functions, and parameters of a model, which are all chosen during training, significantly influence the computation and communication overhead of a cryptographic method, during inference. Based on this observation, we propose SOTERIA -- a training method to construct model architectures that are by-design efficient for private inference. We use neural architecture search algorithms with the dual objective of optimizing the accuracy of the model and the overhead of using cryptographic primitives for secure inference. Given the flexibility of modifying a model during training, we find accurate models that are also efficient for private computation. We select garbled circuits as our underlying cryptographic primitive, due to their expressiveness and efficiency, but this approach can be extended to hybrid multi-party computation settings. We empirically evaluate SOTERIA on MNIST and CIFAR10 datasets, to compare with the prior work. Our results confirm that SOTERIA is indeed effective in balancing performance and accuracy.
The ever-increasing take-up of machine learning techniques requires ever-more application-specific training data. Manually collecting such training data is a tedious and time-consuming process. Data marketplaces represent a compelling alternative, providing an easy way for acquiring data from potential data providers. A key component of such marketplaces is the compensation mechanism for data providers. Classic payoff-allocation methods such as the Shapley value can be vulnerable to data-replication attacks, and are infeasible to compute in the absence of efficient approximation algorithms. To address these challenges, we present an extensive theoretical study on the vulnerabilities of game theoretic payoff-allocation schemes to replication attacks. Our insights apply to a wide range of payoff-allocation schemes, and enable the design of customised replication-robust payoff-allocations. Furthermore, we present a novel efficient sampling algorithm for approximating payoff-allocation schemes based on marginal contributions. In our experiments, we validate the replication-robustness of classic payoff-allocation schemes and new payoff-allocation schemes derived from our theoretical insights. We also demonstrate the efficiency of our proposed sampling algorithm on a wide range of machine learning tasks.
Learning invariant representations has been proposed as a key technique for addressing the domain generalization problem. However, the question of identifying the right conditions for invariance remains unanswered. In this work, we propose a causal interpretation of domain generalization that defines domains as interventions under a data-generating process. Based on a general causal model for data from multiple domains, we show that prior methods for learning an invariant representation optimize for an incorrect objective. We highlight an alternative condition: inputs across domains should have the same representation if they are derived from the same base object. In practice, knowledge about generation of data or objects is not available. Hence we propose an iterative algorithm called MatchDG that approximates base object similarity by using a contrastive loss formulation adapted for multiple domains. We then match inputs that are similar under the resultant representation to build an invariant classifier. We evaluate MatchDG on rotated MNIST, Fashion-MNIST, and PACS datasets and find that it outperforms prior work on out-of-domain accuracy and learns matches that have over 25\% overlap with ground-truth object matches in MNIST and Fashion-MNIST. Code repository can be accessed here: \textit{https://github.com/microsoft/robustdg}
Multi-party machine learning allows several parties to build a joint model to get insights that may not be learnable using only their local data. We consider settings where each party obtains black-box access to the model computed by their mutually agreed-upon algorithm on their joined data. We show that such multi-party computation can cause information leakage between the parties. In particular, a "curious" party can infer the distribution of sensitive attributes in other parties' data with high accuracy. In order to understand and measure the source of leakage, we consider several models of correlation between a sensitive attribute and the rest of the data. Using multiple datasets and machine learning models, we show that leakage occurs even if the sensitive attribute is not included in the training data and has a low correlation with other attributes and the target variable.
This paper aims to enable training and inference of neural networks in a manner that protects the privacy of sensitive data. We propose FALCON - an end-to-end 3-party protocol for fast and secure computation of deep learning algorithms on large networks. FALCON presents three main advantages. It is highly expressive. To the best of our knowledge, it is the first secure framework to support high capacity networks with over a hundred million parameters such as VGG16 as well as the first to support batch normalization, a critical component of deep learning that enables training of complex network architectures such as AlexNet. Next, FALCON guarantees security with abort against malicious adversaries, assuming an honest majority. It ensures that the protocol always completes with correct output for honest participants or aborts when it detects the presence of a malicious adversary. Lastly, FALCON presents new theoretical insights for protocol design that make it highly efficient and allow it to outperform existing secure deep learning solutions. Compared to prior art for private inference, we are about 8x faster than SecureNN (PETS '19) on average and comparable to ABY3 (CCS '18). We are about 16-200x more communication efficient than either of these. For private training, we are about 6x faster than SecureNN, 4.4x faster than ABY3 and about 2-60x more communication efficient. This is the first paper to show via experiments in the WAN setting, that for multi-party machine learning computations over large networks and datasets, compute operations dominate the overall latency, as opposed to the communication.
To continuously improve quality and reflect changes in data, machine learning-based services have to regularly re-train and update their core models. In the setting of language models, we show that a comparative analysis of model snapshots before and after an update can reveal a surprising amount of detailed information about the changes in the data used for training before and after the update. We discuss the privacy implications of our findings, propose mitigation strategies and evaluate their effect.