Initialization Matters: Privacy-Utility Analysis of Overparameterized Neural Networks

• URL: http://arxiv.org/abs/2310.20579v1
• Date: Tue, 31 Oct 2023 16:13:22 GMT
• Title: Initialization Matters: Privacy-Utility Analysis of Overparameterized Neural Networks
• Authors: Jiayuan Ye, Zhenyu Zhu, Fanghui Liu, Reza Shokri, Volkan Cevher
• Abstract summary: We prove a privacy bound for the KL divergence between model distributions on worst-case neighboring datasets. We find that this KL privacy bound is largely determined by the expected squared gradient norm relative to model parameters during training.
• Score: 72.51255282371805
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We analytically investigate how over-parameterization of models in randomized machine learning algorithms impacts the information leakage about their training data. Specifically, we prove a privacy bound for the KL divergence between model distributions on worst-case neighboring datasets, and explore its dependence on the initialization, width, and depth of fully connected neural networks. We find that this KL privacy bound is largely determined by the expected squared gradient norm relative to model parameters during training. Notably, for the special setting of linearized network, our analysis indicates that the squared gradient norm (and therefore the escalation of privacy loss) is tied directly to the per-layer variance of the initialization distribution. By using this analysis, we demonstrate that privacy bound improves with increasing depth under certain initializations (LeCun and Xavier), while degrades with increasing depth under other initializations (He and NTK). Our work reveals a complex interplay between privacy and depth that depends on the chosen initialization distribution. We further prove excess empirical risk bounds under a fixed KL privacy budget, and show that the interplay between privacy utility trade-off and depth is similarly affected by the initialization.

Related papers

• Data Analytics with Differential Privacy [0.0]
  We develop differentially private algorithms to analyze distributed and streaming data. In the distributed model, we consider the particular problem of learning -- in a distributed fashion -- a global model of the data. We offer one of the strongest privacy guarantees for the streaming model, user-level pan-privacy.
  arXiv  Detail & Related papers  (2023-07-20T17:43:29Z)
• Differentially Private Stochastic Gradient Descent with Low-Noise [49.981789906200035]
  Modern machine learning algorithms aim to extract fine-grained information from data to provide accurate predictions, which often conflicts with the goal of privacy protection. This paper addresses the practical and theoretical importance of developing privacy-preserving machine learning algorithms that ensure good performance while preserving privacy.
  arXiv  Detail & Related papers  (2022-09-09T08:54:13Z)
• A Differentially Private Framework for Deep Learning with Convexified Loss Functions [4.059849656394191]
  Differential privacy (DP) has been applied in deep learning for preserving privacy of the underlying training sets. Existing DP practice falls into three categories - objective perturbation, gradient perturbation and output perturbation. We propose a novel output perturbation framework by injecting DP noise into a randomly sampled neuron.
  arXiv  Detail & Related papers  (2022-04-03T11:10:05Z)
• Sensitivity analysis in differentially private machine learning using hybrid automatic differentiation [54.88777449903538]
  We introduce a novel textithybrid automatic differentiation (AD) system for sensitivity analysis. This enables modelling the sensitivity of arbitrary differentiable function compositions, such as the training of neural networks on private data. Our approach can enable the principled reasoning about privacy loss in the setting of data processing.
  arXiv  Detail & Related papers  (2021-07-09T07:19:23Z)
• Understanding Clipping for Federated Learning: Convergence and Client-Level Differential Privacy [67.4471689755097]
  This paper empirically demonstrates that the clipped FedAvg can perform surprisingly well even with substantial data heterogeneity. We provide the convergence analysis of a differential private (DP) FedAvg algorithm and highlight the relationship between clipping bias and the distribution of the clients' updates.
  arXiv  Detail & Related papers  (2021-06-25T14:47:19Z)
• Differential Privacy Dynamics of Langevin Diffusion and Noisy Gradient Descent [10.409652277630132]
  We model the dynamics of privacy loss in Langevin diffusion and extend it to the noisy gradient descent algorithm. We prove that the privacy loss converges exponentially fast.
  arXiv  Detail & Related papers  (2021-02-11T05:49:37Z)
• Robustness Threats of Differential Privacy [70.818129585404]
  We experimentally demonstrate that networks, trained with differential privacy, in some settings might be even more vulnerable in comparison to non-private versions. We study how the main ingredients of differentially private neural networks training, such as gradient clipping and noise addition, affect the robustness of the model.
  arXiv  Detail & Related papers  (2020-12-14T18:59:24Z)
• Privacy-preserving Traffic Flow Prediction: A Federated Learning Approach [61.64006416975458]
  We propose a privacy-preserving machine learning technique named Federated Learning-based Gated Recurrent Unit neural network algorithm (FedGRU) for traffic flow prediction. FedGRU differs from current centralized learning methods and updates universal learning models through a secure parameter aggregation mechanism. It is shown that FedGRU's prediction accuracy is 90.96% higher than the advanced deep learning models.
  arXiv  Detail & Related papers  (2020-03-19T13:07:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.