Random Data Perturbation Research Articles

PyTracer: Automatically Profiling Numerical Instabilities in Python

Numerical stability is a crucial requirement of reliable scientific computing. However, despite the pervasiveness of Python in data science, analyzing large Python programs remains challenging due to the lack of scalable numerical analysis tools available for this language. To fill this gap, we developed PyTracer, a profiler to quantify numerical instability in Python applications. PyTracer transparently instruments Python code to produce numerical traces and visualize them interactively in a Plotly dashboard. We designed PyTracer to be agnostic to numerical noise model, allowing for numerical profiling through Monte-Carlo Arithmetic, random rounding, random data perturbation, or structured noise for a particular application. We illustrate PyTracer's capabilities by testing the numerical stability of key functions in both SciPy and Scikit-learn, two dominant Python libraries for mathematical modeling. Through these evaluations, we demonstrate PyTracer as a scalable, automated, and generic framework for numerical profiling in Python.

Efficient and Privacy-Preserving Speaker Recognition for Cybertwin-Driven 6G

With the introduction of cybertwin, a new approach to represent human or things in the cyberspace, it is foreseeable that vehicles will be able to offer more and more services in the future. Naturally, considering the safety of drivers, speaker recognition will be widely used in vehicle scenarios. Speaker recognition technologies are experiencing increasing popularity due to the unique and indissoluble link between individuals and their voices. However, the coming cybertwin-driven 6G brings speaker recognition technologies unprecedented challenges, especially in preventing the disclosure of voiceprint. To address these challenges, an efficient and privacy-preserving speaker recognition scheme for cybertwin-driven 6G, referred to as NEATEN, is proposed in this article. With NEATEN, the speaker identity can be recognized at multiple security levels without leaking the voiceprint data. More concretely, based on the random projection data perturbation, voiceprint perturbation algorithms in two phases and the corresponding ciphertext-based similarity computation algorithm are proposed. By using these algorithms, our efficient and accurate speaker recognition scheme can be achieved. Orthogonal to the previous works of biometric identification based on the Euclidean distance, NEATEN makes progress on the non-Euclidean distance, such as cosine distance and complicated distance. Detailed analysis shows that NEATEN can resist various known security threats. Experiments conducted on TIMIT and Voxceleb data sets have demonstrated that NEATEN is highly accurate and efficient, and can be flexibly deployed in a real cybertwin-driven 6G vehicle environment.

Privacy Preserving Computation of Scalar Product and Sign of Scalar Product

Scalar product is repeatedly used in a large family of applications. It is a useful building block of many scientific and business applications. Secure scalar product computation is widely used in many privacy-preserving computation domains. In this work, we have proposed privacy preserving protocols for computing scalar product or only its sign. These protocols work on two basic principles - random data perturbation and algebraic methods, and thus avoid standard encryption/decryption overheads. We have designed the protocols to get sign of the scalar product, which can act as a basic building block in privacy preserving database queries over distributed databases and many other applications. We have developed two sets of protocols for the two-party computation with and without using an untrusted third-party. Experimental results on generated random datasets show that the proposed protocols indeed involve low communication overhead.

Protecting Sensitive Data using Multilevel Privacy Preserving Framework Withstanding Linear and Non Linear Attacks

Rapid development of data mining technologies has brought a serious threat to the security of private sensitive information. There is unwanted disclosure of private sensitive information during the process of data collection, data publishing and data mining operations. Developing data mining models without accessing private information as well as without compromising the utility of mining result has become a major concern. Various approaches towards Data Perturbation techniques are limited with manipulation of Gaussian noise to the original data. But Gaussian noise is likely prone to linear attacks. In this paper, we study the effect of adding Gaussian and Laplacian noise as a two step process for random data perturbation. We also consider this effect in multiple trust levels, where the data miners receive their perturbed copies only according to their trust levels. We prove that our solution is vigorous against both linear and non linear attacks. Malicious data miner cannot reconstruct the original data by having access to differently perturbed copies of same data. Our solution thus proves that the combination of Gaussian and Laplacian noise can withstand with both linear and non linear attacks.

Non-Additive Random Data Perturbation for Real World Data

Privacy preservation is the major concern while real datasets are handled. A specific topic- privacy preserving data mining (PPDM), completely deals with data modification but also limits rule loss. Data perturbation is one of the PPDM techniques, which mostly deals with numerical data and concentrates on the statistical analysis of the data. Perturbation is of two types, additive perturbation and multiplicative perturbation, where generated random data is either added or multiplied with the data, which results in a random modified data. In this paper we have proposed a model in which the perturbation is done by randomization, where the data is generated in intervals based on the level of privacy specified by each customer. Our model is proved by applying classification algorithm on the perturbed data set and the accuracy is still maintained the same.

Distributed Noise Generation for Density Estimation Based Clustering without Trusted Third Party

The rapid growth of the Internet provides people with tremendous opportunities for data collection, knowledge discovery and cooperative computation. However, it also brings the problem of sensitive information leakage. Both individuals and enterprises may suffer from the massive data collection and the information retrieval by distrusted parties. In this paper, we propose a privacy-preserving protocol for the distributed kernel density estimation-based clustering. Our scheme applies random data perturbation (RDP) technique and the verifiable secret sharing to solve the security problem of distributed kernel density estimation in [4] which assumed a mediate party to help in the computation.

An Information-Theoretic Approach to Inference Attacks on Random Data Perturbation and a Related Privacy Measure

Random data perturbation (RDP) has been in use for several years in statistical databases and public surveys as a means of providing privacy to individuals while collecting information on groups, and has recently gained popularity as a privacy technique in data mining. This correspondence provides an information-theoretic framework for all inference attacks on RDP. The framework is used to demonstrate the existence of a tight asymptotic lower bound on the number of queries required per bit of entropy for all inference attacks with zero asymptotic error and bounded average power in the query sequence. A privacy measure based on security against inference attacks is proposed.

A cubic-wise balance approach for privacy preservation in data cubes

A data warehouse stores current and historical records consolidated from multiple transactional systems. Securing data warehouses is of ever-increasing interest, especially considering areas where data are sold in pieces to third parties for data mining practices. In this case, existing data warehouse security techniques, such as data access control, may not be easy to enforce and can be ineffective. Instead, this paper proposes a data perturbation based approach, called the cubic-wise balance method, to provide privacy preserving range queries on data cubes in a data warehouse. This approach is motivated by the following observation: analysts are usually interested in summary data rather than individual data values. Indeed, our approach can provide a closely estimated summary data for range queries without providing access to actual individual data values. As demonstrated by our experimental results on APB benchmark data set from the OLAP council, the cubic-wise balance method can achieve both better privacy preservation and better range query accuracy than random data perturbation alternatives.

Security of random data perturbation methods

Statistical databases often use random data perturbation (RDP) methods to protect against disclosure of confidential numerical attributes. One of the key requirements of RDP methods is that they provide the appropriate level of security against snoopers who attempt to obtain information on confidential attributes through statistical inference. In this study, we evaluate the security provided by three methods of perturbation. The results of this study allow the database administrator to select the most effective RDP method that assures adequate protection against disclosure of confidential information.

A modified random perturbation method for database security

The random data perturbation (RDP) method of preserving the privacy of individual records in a statistical database is discussed. In particular, it is shown that if confidential attributes are allowed as query-defining variables, severe biases may result in responses to queries. It is also shown that even if query definition through confidential variables is not allowed, biases can still occur in responses to queries such as those involving proportions or counts. In either case, serious distortions may occur in user statistical analyses. A modified version of RDP is presented, in the form of a query adjustment procedure and specialized perturbation structure which will produce unbiased results.