Nowadays, differential privacy (DP) has become a well-accepted standard for privacy protection, and deep neural networks (DNN) have been immensely successful in machine learning. The combination of these two techniques, i.e., deep learning with differential privacy, promises the privacy-preserving release of high-utility models trained with sensitive data such as medical records. A classic mechanism for this purpose is DP-SGD, which is a differentially private version of the stochastic gradient descent (SGD) optimizer commonly used for DNN training. Subsequent approaches have improved various aspects of the model training process, including noise decay schedule, model architecture, feature engineering, and hyperparameter tuning. However, the core mechanism for enforcing DP in the SGD optimizer remains unchanged ever since the original DP-SGD algorithm, which has increasingly become a fundamental barrier limiting the performance of DP-compliant machine learning solutions. Motivated by this, we propose DPIS, a novel mechanism for differentially private SGD training that can be used as a drop-in replacement of the core optimizer of DP-SGD, with consistent and significant accuracy gains over the latter. The main idea is to employ importance sampling (IS) in each SGD iteration for mini-batch selection, which reduces both sampling variance and the amount of random noise injected to the gradients that is required to satisfy DP. Integrating IS into the complex mathematical machinery of DP-SGD is highly non-trivial. DPIS addresses the challenge through novel mechanism designs, fine-grained privacy analysis, efficiency enhancements, and an adaptive gradient clipping optimization. Extensive experiments on four benchmark datasets, namely MNIST, FMNIST, CIFAR-10 and IMDb, demonstrate the superior effectiveness of DPIS over existing solutions for deep learning with differential privacy.
相關內容
Informed consent has become increasingly salient for data privacy and its regulation. Entities from governments to for-profit companies have addressed concerns about data privacy with policies that enumerate the conditions for personal data storage and transfer. However, increased enumeration of and transparency in data privacy policies has not improved end-users' comprehension of how their data might be used: not only are privacy policies written in legal language that users may struggle to understand, but elements of these policies may compose in such a way that the consequences of the policy are not immediately apparent. We present a framework that uses Answer Set Programming (ASP) -- a type of logic programming -- to formalize privacy policies. Privacy policies thus become constraints on a narrative planning space, allowing end-users to forward-simulate possible consequences of the policy in terms of actors having roles and taking actions in a domain. We demonstrate through the example of the Health Insurance Portability and Accountability Act (HIPAA) how to use the system in various ways, including asking questions about possibilities and identifying which clauses of the law are broken by a given sequence of events.
We investigate the domain of satisfiable formulas in satisfiability modulo theories (SMT), in particular, automatic generation of a multitude of satisfying assignments to such formulas. Despite the long and successful history of SMT in model checking and formal verification, this aspect is relatively under-explored. Prior work exists for generating such assignments, or samples, for Boolean formulas and for quantifier-free first-order formulas involving bit-vectors, arrays, and uninterpreted functions (QF_AUFBV). We propose a new approach that is suitable for a theory T of integer arithmetic and to T with arrays and uninterpreted functions. The approach involves reducing the general sampling problem to a simpler instance of sampling from a set of independent intervals, which can be done efficiently. Such reduction is carried out by expanding a single model - a seed - using top-down propagation of constraints along the original first-order formula.
The performance of differentially private machine learning can be boosted significantly by leveraging the transfer learning capabilities of non-private models pretrained on large public datasets. We critically review this approach. We primarily question whether the use of large Web-scraped datasets should be viewed as differential-privacy-preserving. We caution that publicizing these models pretrained on Web data as "private" could lead to harm and erode the public's trust in differential privacy as a meaningful definition of privacy. Beyond the privacy considerations of using public data, we further question the utility of this paradigm. We scrutinize whether existing machine learning benchmarks are appropriate for measuring the ability of pretrained models to generalize to sensitive domains, which may be poorly represented in public Web data. Finally, we notice that pretraining has been especially impactful for the largest available models -- models sufficiently large to prohibit end users running them on their own devices. Thus, deploying such models today could be a net loss for privacy, as it would require (private) data to be outsourced to a more compute-powerful third party. We conclude by discussing potential paths forward for the field of private learning, as public pretraining becomes more popular and powerful.
Differential privacy (DP) provides a formal privacy guarantee that prevents adversaries with access to machine learning models from extracting information about individual training points. Differentially private stochastic gradient descent (DPSGD) is the most popular training method with differential privacy in image recognition. However, existing DPSGD schemes lead to significant performance degradation, which prevents the application of differential privacy. In this paper, we propose an efficient differential privacy deep learning which accepts a candidate update with a probability that depends both on the update quality and on the number of iterations. Through this random update screening, we make the differentially private gradient descent proceed in the right direction in each iteration, and result in a more accurate model finally. In our experiments, under the same hyperparameters, our scheme achieves test accuracies 98.35%, 87.41% and 60.92% on datasets MNIST, FashionMNIST and CIFAR10, respectively, compared to the state-of-the-art result of 98.12%, 86.33% and 59.34%. Under the freely adjusted hyperparameters, our scheme achieves even higher accuracies, 98.89%, 88.50% and 64.17%. We believe that our method has a great contribution for closing the accuracy gap between private and non-private image classification.
Classical differential private DP-SGD implements individual clipping with random subsampling, which forces a mini-batch SGD approach. We provide a general differential private algorithmic framework that goes beyond DP-SGD and allows any possible first order optimizers (e.g., classical SGD and momentum based SGD approaches) in combination with batch clipping, which clips an aggregate of computed gradients rather than summing clipped gradients (as is done in individual clipping). The framework also admits sampling techniques beyond random subsampling such as shuffling. Our DP analysis follows the $f$-DP approach and introduces a new proof technique which allows us to also analyse group privacy. In particular, for $E$ epochs work and groups of size $g$, we show a $\sqrt{g E}$ DP dependency for batch clipping with shuffling. This is much better than the previously anticipated linear dependency in $g$ and is much better than the previously expected square root dependency on the total number of rounds within $E$ epochs which is generally much more than $\sqrt{E}$.
The problem of adversarial defenses for image classification, where the goal is to robustify a classifier against adversarial examples, is considered. Inspired by the hypothesis that these examples lie beyond the natural image manifold, a novel aDversarIal defenSe with local impliCit functiOns (DISCO) is proposed to remove adversarial perturbations by localized manifold projections. DISCO consumes an adversarial image and a query pixel location and outputs a clean RGB value at the location. It is implemented with an encoder and a local implicit module, where the former produces per-pixel deep features and the latter uses the features in the neighborhood of query pixel for predicting the clean RGB value. Extensive experiments demonstrate that both DISCO and its cascade version outperform prior defenses, regardless of whether the defense is known to the attacker. DISCO is also shown to be data and parameter efficient and to mount defenses that transfers across datasets, classifiers and attacks.
Computational imaging has been revolutionized by compressed sensing algorithms, which offer guaranteed uniqueness, convergence, and stability properties. In recent years, model-based deep learning methods that combine imaging physics with learned regularization priors have been emerging as more powerful alternatives for image recovery. The main focus of this paper is to introduce a memory efficient model-based algorithm with similar theoretical guarantees as CS methods. The proposed iterative algorithm alternates between a gradient descent involving the score function and a conjugate gradient algorithm to encourage data consistency. The score function is modeled as a monotone convolutional neural network. Our analysis shows that the monotone constraint is necessary and sufficient to enforce the uniqueness of the fixed point in arbitrary inverse problems. In addition, it also guarantees the convergence to a fixed point, which is robust to input perturbations. Current algorithms including RED and MoDL are special cases of the proposed algorithm; the proposed theoretical tools enable the optimization of the framework for the deep equilibrium setting. The proposed deep equilibrium formulation is significantly more memory efficient than unrolled methods, which allows us to apply it to 3D or 2D+time problems that current unrolled algorithms cannot handle.
Recent development in the field of explainable artificial intelligence (XAI) has helped improve trust in Machine-Learning-as-a-Service (MLaaS) systems, in which an explanation is provided together with the model prediction in response to each query. However, XAI also opens a door for adversaries to gain insights into the black-box models in MLaaS, thereby making the models more vulnerable to several attacks. For example, feature-based explanations (e.g., SHAP) could expose the top important features that a black-box model focuses on. Such disclosure has been exploited to craft effective backdoor triggers against malware classifiers. To address this trade-off, we introduce a new concept of achieving local differential privacy (LDP) in the explanations, and from that we establish a defense, called XRand, against such attacks. We show that our mechanism restricts the information that the adversary can learn about the top important features, while maintaining the faithfulness of the explanations.
Zero-shot quantization is a promising approach for developing lightweight deep neural networks when data is inaccessible owing to various reasons, including cost and issues related to privacy. By utilizing the learned parameters (statistics) of FP32-pre-trained models, zero-shot quantization schemes focus on generating synthetic data by minimizing the distance between the learned parameters ($\mu$ and $\sigma$) and distributions of intermediate activations. Subsequently, they distill knowledge from the pre-trained model (\textit{teacher}) to the quantized model (\textit{student}) such that the quantized model can be optimized with the synthetic dataset. In general, zero-shot quantization comprises two major elements: synthesizing datasets and quantizing models. However, thus far, zero-shot quantization has primarily been discussed in the context of quantization-aware training methods, which require task-specific losses and long-term optimization as much as retraining. We thus introduce a post-training quantization scheme for zero-shot quantization that produces high-quality quantized networks within a few hours on even half an hour. Furthermore, we propose a framework called \genie~that generates data suited for post-training quantization. With the data synthesized by \genie, we can produce high-quality quantized models without real datasets, which is comparable to few-shot quantization. We also propose a post-training quantization algorithm to enhance the performance of quantized models. By combining them, we can bridge the gap between zero-shot and few-shot quantization while significantly improving the quantization performance compared to that of existing approaches. In other words, we can obtain a unique state-of-the-art zero-shot quantization approach.
Motivated by personalized healthcare and other applications involving sensitive data, we study online exploration in reinforcement learning with differential privacy (DP) constraints. Existing work on this problem established that no-regret learning is possible under joint differential privacy (JDP) and local differential privacy (LDP) but did not provide an algorithm with optimal regret. We close this gap for the JDP case by designing an $\epsilon$-JDP algorithm with a regret of $\widetilde{O}(\sqrt{SAH^2T}+S^2AH^3/\epsilon)$ which matches the information-theoretic lower bound of non-private learning for all choices of $\epsilon> S^{1.5}A^{0.5} H^2/\sqrt{T}$. In the above, $S$, $A$ denote the number of states and actions, $H$ denotes the planning horizon, and $T$ is the number of steps. To the best of our knowledge, this is the first private RL algorithm that achieves \emph{privacy for free} asymptotically as $T\rightarrow \infty$. Our techniques -- which could be of independent interest -- include privately releasing Bernstein-type exploration bonuses and an improved method for releasing visitation statistics. The same techniques also imply a slightly improved regret bound for the LDP case.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191