Differentially private SGD (DPSGD) has recently shown promise in deep learning. However, compared to non-private SGD, the DPSGD algorithm places computational overheads that can undo the benefit of batching in GPUs. Micro-batching is a common method to alleviate this and is fully supported in the TensorFlow Privacy library (TFDP). However, it degrades accuracy. We propose NanoBatch Privacy, a lightweight add-on to TFDP to be used on Graphcore IPUs by leveraging batch size of 1 (without microbatching) and gradient accumulation. This allows us to achieve large total batch sizes with minimal impacts to throughput. Second, we illustrate using Cifar-10 how larger batch sizes are not necessarily optimal from a privacy versus utility perspective. On ImageNet, we achieve more than 15x speedup over TFDP versus 8x A100s and significant speedups even across libraries such as Opacus. We also provide two extensions: 1) DPSGD for pipelined models and 2) per-layer clipping that is 15x faster than the Opacus implementation on 8x A100s. Finally as an application case study, we apply NanoBatch training for use on private Covid-19 chest CT prediction.
相關內容
We identify a new class of vulnerabilities in implementations of differential privacy. Specifically, they arise when computing basic statistics such as sums, thanks to discrepancies between the implemented arithmetic using finite data types (namely, ints or floats) and idealized arithmetic over the reals or integers. These discrepancies cause the sensitivity of the implemented statistics (i.e., how much one individual's data can affect the result) to be much higher than the sensitivity we expect. Consequently, essentially all differential privacy libraries fail to introduce enough noise to hide individual-level information as required by differential privacy, and we show that this may be exploited in realistic attacks on differentially private query systems. In addition to presenting these vulnerabilities, we also provide a number of solutions, which modify or constrain the way in which the sum is implemented in order to recover the idealized or near-idealized bounds on sensitivity.
Federated learning enables resource-constrained edge compute devices, such as mobile phones and IoT devices, to learn a shared model for prediction, while keeping the training data local. This decentralized approach to train models provides privacy, security, regulatory and economic benefits. In this work, we focus on the statistical challenge of federated learning when local data is non-IID. We first show that the accuracy of federated learning reduces significantly, by up to 55% for neural networks trained for highly skewed non-IID data, where each client device trains only on a single class of data. We further show that this accuracy reduction can be explained by the weight divergence, which can be quantified by the earth mover's distance (EMD) between the distribution over classes on each device and the population distribution. As a solution, we propose a strategy to improve training on non-IID data by creating a small subset of data which is globally shared between all the edge devices. Experiments show that accuracy can be increased by 30% for the CIFAR-10 dataset with only 5% globally shared data.
Federated learning (FL) is a recently developed area of machine learning, in which the private data of a large number of distributed clients is used to develop a global model under the coordination of a central server without explicitly exposing the data. The standard FL strategy has a number of significant bottlenecks including large communication requirements and high impact on the clients' resources. Several strategies have been described in the literature trying to address these issues. In this paper, a novel scheme based on the notion of "model growing" is proposed. Initially, the server deploys a small model of low complexity, which is trained to capture the data complexity during the initial set of rounds. When the performance of such a model saturates, the server switches to a larger model with the help of function-preserving transformations. The model complexity increases as more data is processed by the clients, and the overall process continues until the desired performance is achieved. Therefore, the most complex model is broadcast only at the final stage in our approach resulting in substantial reduction in communication cost and client computational requirements. The proposed approach is tested extensively on three standard benchmarks and is shown to achieve substantial reduction in communication and client computation while achieving comparable accuracy when compared to the current most effective strategies.
In this paper, we study the problem of learning Graph Neural Networks (GNNs) with Differential Privacy (DP). We propose a novel differentially private GNN based on Aggregation Perturbation (GAP), which adds stochastic noise to the GNN's aggregation function to statistically obfuscate the presence of a single edge (edge-level privacy) or a single node and all its adjacent edges (node-level privacy). Tailored to the specifics of private learning, GAP's new architecture is composed of three separate modules: (i) the encoder module, where we learn private node embeddings without relying on the edge information; (ii) the aggregation module, where we compute noisy aggregated node embeddings based on the graph structure; and (iii) the classification module, where we train a neural network on the private aggregations for node classification without further querying the graph edges. GAP's major advantage over previous approaches is that it can benefit from multi-hop neighborhood aggregations, and guarantees both edge-level and node-level DP not only for training, but also at inference with no additional costs beyond the training's privacy budget. We analyze GAP's formal privacy guarantees using R\'enyi DP and conduct empirical experiments over three real-world graph datasets. We demonstrate that GAP offers significantly better accuracy-privacy trade-offs than state-of-the-art DP-GNN approaches and naive MLP-based baselines.
An interactive mechanism is an algorithm that stores a data set and answers adaptively chosen queries to it. The mechanism is called differentially private, if any adversary cannot distinguish whether a specific individual is in the data set by interacting with the mechanism. We study composition properties of differential privacy in concurrent compositions. In this setting, an adversary interacts with k interactive mechanisms in parallel and can interleave its queries to the mechanisms arbitrarily. Previously, [Vadhan and Wang, TCC 2021] proved an optimal concurrent composition theorem for pure-differential privacy. We significantly generalize and extend their results. Namely, we prove optimal parallel composition properties for several major notions of differential privacy in the literature, including approximate DP, R\'enyi DP, and zero-concentrated DP. Our results demonstrate that the adversary gains no advantage by interleaving its queries to independently running mechanisms. Hence, interactivity is a feature that differential privacy grants us for free.
Numerous studies in the literature have already shown the potential of biometrics on mobile devices for authentication purposes. However, it has been shown that, the learning processes associated to biometric systems might expose sensitive personal information about the subjects. This study proposes GaitPrivacyON, a novel mobile gait biometrics verification approach that provides accurate authentication results while preserving the sensitive information of the subject. It comprises two modules: i) a convolutional Autoencoder that transforms attributes of the biometric raw data, such as the gender or the activity being performed, into a new privacy-preserving representation; and ii) a mobile gait verification system based on the combination of Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with a Siamese architecture. The main advantage of GaitPrivacyON is that the first module (convolutional Autoencoder) is trained in an unsupervised way, without specifying the sensitive attributes of the subject to protect. The experimental results achieved using two popular databases (MotionSense and MobiAct) suggest the potential of GaitPrivacyON to significantly improve the privacy of the subject while keeping user authentication results higher than 99% Area Under the Curve (AUC). To the best of our knowledge, this is the first mobile gait verification approach that considers privacy-preserving methods trained in an unsupervised way.
Federated Learning (FL), a distributed machine learning paradigm, has been adapted to mitigate privacy concerns for customers. Despite their appeal, there are various inference attacks that can exploit shared-plaintext model updates to embed traces of customer private information, leading to serious privacy concerns. To alleviate this privacy issue, cryptographic techniques such as Secure Multi-Party Computation and Homomorphic Encryption have been used for privacy-preserving FL. However, such security issues in privacy-preserving FL are poorly elucidated and underexplored. This work is the first attempt to elucidate the triviality of performing model corruption attacks on privacy-preserving FL based on lightweight secret sharing. We consider scenarios in which model updates are quantized to reduce communication overhead in this case, where an adversary can simply provide local parameters outside the legal range to corrupt the model. We then propose the MUD-PQFed protocol, which can precisely detect malicious clients performing attacks and enforce fair penalties. By removing the contributions of detected malicious clients, the global model utility is preserved to be comparable to the baseline global model without the attack. Extensive experiments validate effectiveness in maintaining baseline accuracy and detecting malicious clients in a fine-grained manner
Federated Learning (FL) is a technique to train models using data distributed across devices. Differential Privacy (DP) provides a formal privacy guarantee for sensitive data. Our goal is to train a large neural network language model (NNLM) on compute-constrained devices while preserving privacy using FL and DP. However, the DP-noise introduced to the model increases as the model size grows, which often prevents convergence. We propose Partial Embedding Updates (PEU), a novel technique to decrease noise by decreasing payload size. Furthermore, we adopt Low Rank Adaptation (LoRA) and Noise Contrastive Estimation (NCE) to reduce the memory demands of large models on compute-constrained devices. This combination of techniques makes it possible to train large-vocabulary language models while preserving accuracy and privacy.
As data are increasingly being stored in different silos and societies becoming more aware of data privacy issues, the traditional centralized training of artificial intelligence (AI) models is facing efficiency and privacy challenges. Recently, federated learning (FL) has emerged as an alternative solution and continue to thrive in this new reality. Existing FL protocol design has been shown to be vulnerable to adversaries within or outside of the system, compromising data privacy and system robustness. Besides training powerful global models, it is of paramount importance to design FL systems that have privacy guarantees and are resistant to different types of adversaries. In this paper, we conduct the first comprehensive survey on this topic. Through a concise introduction to the concept of FL, and a unique taxonomy covering: 1) threat models; 2) poisoning attacks and defenses against robustness; 3) inference attacks and defenses against privacy, we provide an accessible review of this important topic. We highlight the intuitions, key techniques as well as fundamental assumptions adopted by various attacks and defenses. Finally, we discuss promising future research directions towards robust and privacy-preserving federated learning.
Transfer learning aims at improving the performance of target learners on target domains by transferring the knowledge contained in different but related source domains. In this way, the dependence on a large number of target domain data can be reduced for constructing target learners. Due to the wide application prospects, transfer learning has become a popular and promising area in machine learning. Although there are already some valuable and impressive surveys on transfer learning, these surveys introduce approaches in a relatively isolated way and lack the recent advances in transfer learning. As the rapid expansion of the transfer learning area, it is both necessary and challenging to comprehensively review the relevant studies. This survey attempts to connect and systematize the existing transfer learning researches, as well as to summarize and interpret the mechanisms and the strategies in a comprehensive way, which may help readers have a better understanding of the current research status and ideas. Different from previous surveys, this survey paper reviews over forty representative transfer learning approaches from the perspectives of data and model. The applications of transfer learning are also briefly introduced. In order to show the performance of different transfer learning models, twenty representative transfer learning models are used for experiments. The models are performed on three different datasets, i.e., Amazon Reviews, Reuters-21578, and Office-31. And the experimental results demonstrate the importance of selecting appropriate transfer learning models for different applications in practice.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191