arXiv subjects

Florian Kerschbaum

Publications and source records attributed to Florian Kerschbaum.

At least 55 records · Page 3Linked to original sources

arXiv2022

Assessing Differentially Private Variational Autoencoders under Membership Inference

We present an approach to quantify and compare the privacy-accuracy trade-off for differentially private Variational Autoencoders. Our work complements previous work in two aspects. First, we evaluate the the strong reconstruction MI attack against Variational Autoencoders under differential privacy. Second, we address the data scientist's challenge of setting privacy parameter epsilon, which steers the differential privacy strength and thus also the privacy-accuracy trade-off. In our experimental study we consider image and time series data, and three local and central differential privacy mechanisms. We find that the privacy-accuracy trade-offs strongly depend on the dataset and model architecture. We do rarely observe favorable privacy-accuracy trade-off for Variational Autoencoders, and identify a case where LDP outperforms CDP.

cs.CR↗

arXiv2022

Constant-weight PIR: Single-round Keyword PIR via Constant-weight Equality Operators

Equality operators are an essential building block in tasks over secure computation such as private information retrieval. In private information retrieval (PIR), a user queries a database such that the server does not learn which element is queried. In this work, we propose \emph{equality operators for constant-weight codewords}. A constant-weight code is a collection of codewords that share the same Hamming weight. Constant-weight equality operators have a multiplicative depth that depends only on the Hamming weight of the code, not the bit-length of the elements. In our experiments, we show how these equality operators are up to 10 times faster than existing equality operators. Furthermore, we propose PIR using the constant-weight equality operator or \emph{constant-weight PIR}, which is a PIR protocol using an approach previously deemed impractical. We show that for private retrieval of large, streaming data, constant-weight PIR has a smaller communication complexity and lower runtime compared to SEALPIR and MulPIR, respectively, which are two state-of-the-art solutions for PIR. Moreover, we show how constant-weight PIR can be extended to keyword PIR. In keyword PIR, the desired element is retrieved by a unique identifier pertaining to the sought item, e.g., the name of a file. Previous solutions to keyword PIR require one or multiple rounds of communication to reduce the problem to normal PIR. We show that constant-weight PIR is the first practical single-round solution to single-server keyword PIR.

cs.CR↗

arXiv2021

Generalization Techniques Empirically Outperform Differential Privacy against Membership Inference

Differentially private training algorithms provide protection against one of the most popular attacks in machine learning: the membership inference attack. However, these privacy algorithms incur a loss of the model's classification accuracy, therefore creating a privacy-utility trade-off. The amount of noise that differential privacy requires to provide strong theoretical protection guarantees in deep learning typically renders the models unusable, but authors have observed that even lower noise levels provide acceptable empirical protection against existing membership inference attacks. In this work, we look for alternatives to differential privacy towards empirically protecting against membership inference attacks. We study the protection that simply following good machine learning practices (not designed with privacy in mind) offers against membership inference. We evaluate the performance of state-of-the-art techniques, such as pre-training and sharpness-aware minimization, alone and with differentially private training algorithms, and find that, when using early stopping, the algorithms without differential privacy can provide both higher utility and higher privacy than their differentially private counterparts. These findings challenge the belief that differential privacy is a good defense to protect against existing membership inference attacks

cs.CR↗

arXiv2021

SoK: How Robust is Image Classification Deep Neural Network Watermarking? (Extended Version)

Deep Neural Network (DNN) watermarking is a method for provenance verification of DNN models. Watermarking should be robust against watermark removal attacks that derive a surrogate model that evades provenance verification. Many watermarking schemes that claim robustness have been proposed, but their robustness is only validated in isolation against a relatively small set of attacks. There is no systematic, empirical evaluation of these claims against a common, comprehensive set of removal attacks. This uncertainty about a watermarking scheme's robustness causes difficulty to trust their deployment in practice. In this paper, we evaluate whether recently proposed watermarking schemes that claim robustness are robust against a large set of removal attacks. We survey methods from the literature that (i) are known removal attacks, (ii) derive surrogate models but have not been evaluated as removal attacks, and (iii) novel removal attacks. Weight shifting and smooth retraining are novel removal attacks adapted to the DNN watermarking schemes surveyed in this paper. We propose taxonomies for watermarking schemes and removal attacks. Our empirical evaluation includes an ablation study over sets of parameters for each attack and watermarking scheme on the CIFAR-10 and ImageNet datasets. Surprisingly, none of the surveyed watermarking schemes is robust in practice. We find that schemes fail to withstand adaptive attacks and known methods for deriving surrogate models that have not been evaluated as removal attacks. This points to intrinsic flaws in how robustness is currently evaluated. We show that watermarking schemes need to be evaluated against a more extensive set of removal attacks with a more realistic adversary model. Our source code and a complete dataset of evaluation results are publicly available, which allows to independently verify our conclusions.

cs.CR↗

arXiv2021

Quantifying identifiability to choose and audit $ε$ in differentially private deep learning

Differential privacy allows bounding the influence that training data records have on a machine learning model. To use differential privacy in machine learning, data scientists must choose privacy parameters $(ε,δ)$. Choosing meaningful privacy parameters is key, since models trained with weak privacy parameters might result in excessive privacy leakage, while strong privacy parameters might overly degrade model utility. However, privacy parameter values are difficult to choose for two main reasons. First, the theoretical upper bound on privacy loss $(ε,δ)$ might be loose, depending on the chosen sensitivity and data distribution of practical datasets. Second, legal requirements and societal norms for anonymization often refer to individual identifiability, to which $(ε,δ)$ are only indirectly related. We transform $(ε,δ)$ to a bound on the Bayesian posterior belief of the adversary assumed by differential privacy concerning the presence of any record in the training dataset. The bound holds for multidimensional queries under composition, and we show that it can be tight in practice. Furthermore, we derive an identifiability bound, which relates the adversary assumed in differential privacy to previous work on membership inference adversaries. We formulate an implementation of this differential privacy adversary that allows data scientists to audit model training and compute empirical identifiability scores and empirical $(ε,δ)$.

cs.CR↗

arXiv2021

Equi-Joins Over Encrypted Data for Series of Queries

Encryption provides a method to protect data outsourced to a DBMS provider, e.g., in the cloud. However, performing database operations over encrypted data requires specialized encryption schemes that carefully balance security and performance. In this paper, we present a new encryption scheme that can efficiently perform equi-joins over encrypted data with better security than the state-of-the-art. In particular, our encryption scheme reduces the leakage to equality of rows that match a selection criterion and only reveals the transitive closure of the sum of the leakages of each query in a series of queries. Our encryption scheme is provable secure. We implemented our encryption scheme and evaluated it over a dataset from the TPC-H benchmark.

cs.CR↗

arXiv2021

PCOR: Private Contextual Outlier Release via Differentially Private Search

Outlier detection plays a significant role in various real world applications such as intrusion, malfunction, and fraud detection. Traditionally, outlier detection techniques are applied to find outliers in the context of the whole dataset. However, this practice neglects contextual outliers, that are not outliers in the whole dataset but in some specific neighborhoods. Contextual outliers are particularly important in data exploration and targeted anomaly explanation and diagnosis. In these scenarios, the data owner computes the following information: i) The attributes that contribute to the abnormality of an outlier (metric), ii) Contextual description of the outlier's neighborhoods (context), and iii) The utility score of the context, e.g. its strength in showing the outlier's significance, or in relation to a particular explanation for the outlier. However, revealing the outlier's context leaks information about the other individuals in the population as well, violating their privacy. We address the issue of population privacy violations in this paper, and propose a solution for the two main challenges. In this setting, the data owner is required to release a valid context for the queried record, i.e. a context in which the record is an outlier. Hence, the first major challenge is that the privacy technique must preserve the validity of the context for each record. We propose techniques to protect the privacy of individuals through a relaxed notion of differential privacy to satisfy this requirement. The second major challenge is applying the proposed techniques efficiently, as they impose intensive computation to the base algorithm. To overcome this challenge, we propose a graph structure to map the contexts to, and introduce differentially private graph search algorithms as efficient solutions for the computation problem caused by differential privacy techniques.

cs.DB↗

arXiv2021

Obfuscated Access and Search Patterns in Searchable Encryption

Searchable Symmetric Encryption (SSE) allows a data owner to securely outsource its encrypted data to a cloud server while maintaining the ability to search over it and retrieve matched documents. Most existing SSE schemes leak which documents are accessed per query, i.e., the so-called access pattern, and thus are vulnerable to attacks that can recover the database or the queried keywords. Current techniques that fully hide access patterns, such as ORAM or PIR, suffer from heavy communication or computational costs, and are not designed with search capabilities in mind. Recently, Chen et al. (INFOCOM'18) proposed an obfuscation framework for SSE that protects the access pattern in a differentially private way with a reasonable utility cost. However, this scheme leaks the so-called search pattern, i.e., how many times a certain query is performed. This leakage makes the proposal vulnerable to certain database and query recovery attacks. In this paper, we propose OSSE (Obfuscated SSE), an SSE scheme that obfuscates the access pattern independently for each query performed. This in turn hides the search pattern and makes our scheme resistant against attacks that rely on this leakage. Under certain reasonable assumptions, our scheme has smaller communication overhead than ORAM-based SSE. Furthermore, our scheme works in a single communication round and requires very small constant client-side storage. Our empirical evaluation shows that OSSE is highly effective at protecting against different query recovery attacks while keeping a reasonable utility level. Our protocol provides significantly more protection than the proposal by Chen et al.~against some state-of-the-art attacks, which demonstrates the importance of hiding search patterns in designing effective privacy-preserving SSE schemes.

cs.CR↗

arXiv2021

RIGA: Covert and Robust White-Box Watermarking of Deep Neural Networks

Watermarking of deep neural networks (DNN) can enable their tracing once released by a data owner. In this paper, we generalize white-box watermarking algorithms for DNNs, where the data owner needs white-box access to the model to extract the watermark. White-box watermarking algorithms have the advantage that they do not impact the accuracy of the watermarked model. We propose Robust whIte-box GAn watermarking (RIGA), a novel white-box watermarking algorithm that uses adversarial training. Our extensive experiments demonstrate that the proposed watermarking algorithm not only does not impact accuracy, but also significantly improves the covertness and robustness over the current state-of-art.

cs.CR↗

arXiv2021

Deep Neural Network Fingerprinting by Conferrable Adversarial Examples

In Machine Learning as a Service, a provider trains a deep neural network and gives many users access. The hosted (source) model is susceptible to model stealing attacks, where an adversary derives a surrogate model from API access to the source model. For post hoc detection of such attacks, the provider needs a robust method to determine whether a suspect model is a surrogate of their model. We propose a fingerprinting method for deep neural network classifiers that extracts a set of inputs from the source model so that only surrogates agree with the source model on the classification of such inputs. These inputs are a subclass of transferable adversarial examples which we call conferrable adversarial examples that exclusively transfer with a target label from a source model to its surrogates. We propose a new method to generate these conferrable adversarial examples. We present an extensive study on the irremovability of our fingerprint against fine-tuning, weight pruning, retraining, retraining with different architectures, three model extraction attacks from related work, transfer learning, adversarial training, and two new adaptive attacks. Our fingerprint is robust against distillation, related model extraction attacks, and even transfer learning when the attacker has no access to the model provider's dataset. Our fingerprint is the first method that reaches a ROC AUC of 1.0 in verifying surrogates, compared to a ROC AUC of 0.63 by previous fingerprints.

cs.LG↗

arXiv2020

Efficient Oblivious Database Joins

A major algorithmic challenge in designing applications intended for secure remote execution is ensuring that they are oblivious to their inputs, in the sense that their memory access patterns do not leak sensitive information to the server. This problem is particularly relevant to cloud databases that wish to allow queries over the client's encrypted data. One of the major obstacles to such a goal is the join operator, which is non-trivial to implement obliviously without resorting to generic but inefficient solutions like Oblivious RAM (ORAM). We present an oblivious algorithm for equi-joins which (up to a logarithmic factor) matches the optimal $O(n\log n)$ complexity of the standard non-secure sort-merge join (on inputs producing $O(n)$ outputs). We do not use use expensive primitives like ORAM or rely on unrealistic hardware or security assumptions. Our approach, which is based on sorting networks and novel provably-oblivious constructions, is conceptually simple, easily verifiable, and very efficient in practice. Its data-independent algorithmic structure makes it secure in various different settings for remote computation, even in those that are known to be vulnerable to certain side-channel attacks (such as Intel SGX) or with strict requirements for low circuit complexity (like secure multiparty computation). We confirm that our approach is easily realizable through a compact implementation which matches our expectations for performance and is shown, both formally and empirically, to possess the desired security characteristics.

cs.DB↗

arXiv2020

Hiding the Access Pattern is Not Enough: Exploiting Search Pattern Leakage in Searchable Encryption

Recent Searchable Symmetric Encryption (SSE) schemes enable secure searching over an encrypted database stored in a server while limiting the information leaked to the server. These schemes focus on hiding the access pattern, which refers to the set of documents that match the client's queries. This provides protection against current attacks that largely depend on this leakage to succeed. However, most SSE constructions also leak whether or not two queries aim for the same keyword, also called the search pattern. In this work, we show that search pattern leakage can severely undermine current SSE defenses. We propose an attack that leverages both access and search pattern leakage, as well as some background and query distribution information, to recover the keywords of the queries performed by the client. Our attack follows a maximum likelihood estimation approach, and is easy to adapt against SSE defenses that obfuscate the access pattern. We empirically show that our attack is efficient, it outperforms other proposed attacks, and it completely thwarts two out of the three defenses we evaluate it against, even when these defenses are set to high privacy regimes. These findings highlight that hiding the search pattern, a feature that most constructions are lacking, is key towards providing practical privacy guarantees in SSE.

cs.CR↗

arXiv2020

Assessing differentially private deep learning with Membership Inference

Attacks that aim to identify the training data of public neural networks represent a severe threat to the privacy of individuals participating in the training data set. A possible protection is offered by anonymization of the training data or training function with differential privacy. However, data scientists can choose between local and central differential privacy and need to select meaningful privacy parameters $ε$ which is challenging for non-privacy experts. We empirically compare local and central differential privacy mechanisms under white- and black-box membership inference to evaluate their relative privacy-accuracy trade-offs. We experiment with several datasets and show that this trade-off is similar for both types of mechanisms. This suggests that local differential privacy is a sound alternative to central differential privacy for differentially private deep learning, since small $ε$ in central differential privacy and large $ε$ in local differential privacy result in similar membership inference attack risk.

cs.CR↗

arXiv2020

EncDBDB: Searchable Encrypted, Fast, Compressed, In-Memory Database using Enclaves

Data confidentiality is an important requirement for clients when outsourcing databases to the cloud. Trusted execution environments, such as Intel SGX, offer an efficient, hardware-based solution to this cryptographic problem. Existing solutions are not optimized for column-oriented, in-memory databases and pose impractical memory requirements on the enclave. We present EncDBDB, a novel approach for client-controlled encryption of a column-oriented, in-memory databases allowing range searches using an enclave. EncDBDB offers nine encrypted dictionaries, which provide different security, performance and storage efficiency tradeoffs for the data. It is especially suited for complex, read-oriented, analytic queries, e.g., as present in data warehouses. The computational overhead compared to plaintext processing is within a millisecond even for databases with millions of entries and the leakage is limited. Compressed encrypted data requires less space than a corresponding plaintext column. Furthermore, the resulting code - and data - in the enclave is very small reducing the potential for security-relevant implementation errors and side-channel leakages.

cs.CR↗

arXiv2019

On the Robustness of the Backdoor-based Watermarking in Deep Neural Networks

Obtaining the state of the art performance of deep learning models imposes a high cost to model generators, due to the tedious data preparation and the substantial processing requirements. To protect the model from unauthorized re-distribution, watermarking approaches have been introduced in the past couple of years. We investigate the robustness and reliability of state-of-the-art deep neural network watermarking schemes. We focus on backdoor-based watermarking and propose two -- a black-box and a white-box -- attacks that remove the watermark. Our black-box attack steals the model and removes the watermark with minimum requirements; it just relies on public unlabeled data and a black-box access to the classification label. It does not need classification confidences or access to the model's sensitive information such as the training data set, the trigger set or the model parameters. The white-box attack, proposes an efficient watermark removal when the parameters of the marked model are available; our white-box attack does not require access to the labeled data or the trigger set and improves the runtime of the black-box attack up to seventeen times. We as well prove the security inadequacy of the backdoor-based watermarking in keeping the watermark undetectable by proposing an attack that detects whether a model contains a watermark. Our attacks show that a recipient of a marked model can remove a backdoor-based watermark with significantly less effort than training a new model and some other techniques are needed to protect against re-distribution by a motivated attacker.

cs.LG↗

arXiv2019

Secure Computation of the kth-Ranked Element in a Star Network

We consider the problem of securely computing the kth-ranked element in a sequence of n private integers distributed among n parties. The kth-ranked element (e.g., minimum, maximum, median) is of particular interest in benchmarking, which allows a company to compare its own key performance indicator to the statistics of its peer group. The individual integers are sensitive data, yet the kth-ranked element is of mutual interest to the parties. Previous secure computation protocols for the kth-ranked element require a communication channel between each pair of parties. They do not scale to a large number of parties as they are highly interactive resulting in longer delays. Moreover, they are difficult to deploy as special arrangements are required between each pair of parties to establish a secure connection. A server model naturally fits with the client-server architecture of Internet applications in which clients are connected to the server and not to other clients. It can simplify secure computation by reducing the number of rounds, and as a result, improve its performance and scalability. In this model, there are communication channels only between each client and the server, while only clients provide inputs to the computation. Hence, it is a centralized communication pattern, i.e., a star network. We propose different approaches for privately computing the kth-ranked element in the server model, using either garbled circuits or threshold homomorphic encryption. Our schemes have a constant number of rounds and can compute the kth-ranked element within seconds for up to 50 clients in a WAN.

cs.CR↗

arXiv2019

Non-Interactive Private Decision Tree Evaluation

Decision trees are a powerful prediction model with many applications in statistics, data mining, and machine learning. In some settings, the model and the data to be classified may contain sensitive information belonging to different parties. In this paper, we, therefore, address the problem of privately evaluating a decision tree on private data. This scenario consists of a server holding a private decision tree model and a client interested in classifying its private attribute vector using the server's private model. The goal of the computation is to obtain the classification while preserving the privacy of both - the decision tree and the client input. After the computation, the classification result is revealed only to the client, and nothing else is revealed neither to the client nor to the server. Existing privacy-preserving protocols that address this problem use or combine different generic secure multiparty computation approaches resulting in several interactions between the client and the server. Our goal is to design and implement a novel client-server protocol that delegates the complete tree evaluation to the server while preserving privacy and reducing the overhead. The idea is to use fully (somewhat) homomorphic encryption and evaluate the tree on ciphertexts encrypted under the client's public key. However, since current somewhat homomorphic encryption schemes have high overhead, we combine efficient data representations with different algorithmic optimizations to keep the computational overhead and the communication cost low. As a result, we are able to provide the first non-interactive protocol, that allows the client to delegate the evaluation to the server by sending an encrypted input and receiving only the encryption of the result. Our scheme has only one round and can evaluate a complete tree of depth 10 within seconds.

cs.CR↗

arXiv2018

SynTF: Synthetic and Differentially Private Term Frequency Vectors for Privacy-Preserving Text Mining

Text mining and information retrieval techniques have been developed to assist us with analyzing, organizing and retrieving documents with the help of computers. In many cases, it is desirable that the authors of such documents remain anonymous: Search logs can reveal sensitive details about a user, critical articles or messages about a company or government might have severe or fatal consequences for a critic, and negative feedback in customer surveys might negatively impact business relations if they are identified. Simply removing personally identifying information from a document is, however, insufficient to protect the writer's identity: Given some reference texts of suspect authors, so-called authorship attribution methods can reidentfy the author from the text itself. One of the most prominent models to represent documents in many common text mining and information retrieval tasks is the vector space model where each document is represented as a vector, typically containing its term frequencies or related quantities. We therefore propose an automated text anonymization approach that produces synthetic term frequency vectors for the input documents that can be used in lieu of the original vectors. We evaluate our method on an exemplary text classification task and demonstrate that it only has a low impact on its accuracy. In contrast, we show that our method strongly affects authorship attribution techniques to the level that they become infeasible with a much stronger decline in accuracy. Other than previous authorship obfuscation methods, our approach is the first that fulfills differential privacy and hence comes with a provable plausible deniability guarantee.

cs.CR↗