arXiv subjects

Florian Kerschbaum

Publications and source records attributed to Florian Kerschbaum.

At least 37 records · Page 2Linked to original sources

arXiv2024

PEPSI: Practically Efficient Private Set Intersection in the Unbalanced Setting

Two parties with private data sets can find shared elements using a Private Set Intersection (PSI) protocol without revealing any information beyond the intersection. Circuit PSI protocols privately compute an arbitrary function of the intersection - such as its cardinality, and are often employed in an unbalanced setting where one party has more data than the other. Existing protocols are either computationally inefficient or require extensive server-client communication on the order of the larger set. We introduce Practically Efficient PSI or PEPSI, a non-interactive solution where only the client sends its encrypted data. PEPSI can process an intersection of 1024 client items with a million server items in under a second, using less than 5 MB of communication. Our work is over 4 orders of magnitude faster than an existing non-interactive circuit PSI protocol and requires only 10% of the communication. It is also up to 20 times faster than the work of Ion et al., which computes a limited set of functions and has communication costs proportional to the larger set. Our work is the first to demonstrate that non-interactive circuit PSI can be practically applied in an unbalanced setting.

cs.CR↗

arXiv2024

HE is all you need: Compressing FHE Ciphertexts using Additive HE

Homomorphic Encryption (HE) is a commonly used tool for building privacy-preserving applications. However, in scenarios with many clients and high-latency networks, communication costs due to large ciphertext sizes are the bottleneck. In this paper, we present a new compression technique that uses an additive homomorphic encryption scheme with small ciphertexts to compress large homomorphic ciphertexts based on Learning with Errors (LWE). Our technique exploits the linear step in the decryption of such ciphertexts to delegate part of the decryption to the server. We achieve compression ratios up to 90% which only requires a small compression key. By compressing multiple ciphertexts simultaneously, we can over 99\% compression rate. Our compression technique can be readily applied to applications which transmit LWE ciphertexts from the server to the client as the response to a query. Furthermore, we apply our technique to private information retrieval (PIR) where a client accesses a database without revealing its query. Using our compression technique, we propose ZipPIR, a PIR protocol which achieves the lowest overall communication cost among all protocols in the literature. ZipPIR does not require any communication with the client in the preprocessing phase, making it a great solution for use cases of PIR with ephemeral clients or high-latency networks.

cs.CR↗

arXiv2024

Fast and Private Inference of Deep Neural Networks by Co-designing Activation Functions

Machine Learning as a Service (MLaaS) is an increasingly popular design where a company with abundant computing resources trains a deep neural network and offers query access for tasks like image classification. The challenge with this design is that MLaaS requires the client to reveal their potentially sensitive queries to the company hosting the model. Multi-party computation (MPC) protects the client's data by allowing encrypted inferences. However, current approaches suffer from prohibitively large inference times. The inference time bottleneck in MPC is the evaluation of non-linear layers such as ReLU activation functions. Motivated by the success of previous work co-designing machine learning and MPC, we develop an activation function co-design. We replace all ReLUs with a polynomial approximation and evaluate them with single-round MPC protocols, which give state-of-the-art inference times in wide-area networks. Furthermore, to address the accuracy issues previously encountered with polynomial activations, we propose a novel training algorithm that gives accuracy competitive with plaintext models. Our evaluation shows between $3$ and $110\times$ speedups in inference time on large models with up to $23$ million parameters while maintaining competitive inference accuracy.

cs.CR↗

arXiv2024

SoK: Analyzing Adversarial Examples: A Framework to Study Adversary Knowledge

Adversarial examples are malicious inputs to machine learning models that trigger a misclassification. This type of attack has been studied for close to a decade, and we find that there is a lack of study and formalization of adversary knowledge when mounting attacks. This has yielded a complex space of attack research with hard-to-compare threat models and attacks. We focus on the image classification domain and provide a theoretical framework to study adversary knowledge inspired by work in order theory. We present an adversarial example game, inspired by cryptographic games, to standardize attacks. We survey recent attacks in the image classification domain and classify their adversary's knowledge in our framework. From this systematization, we compile results that both confirm existing beliefs about adversary knowledge, such as the potency of information about the attacked model as well as allow us to derive new conclusions on the difficulty associated with the white-box and transferable threat models, for example, that transferable attacks might not be as difficult as previously thought.

cs.LG↗

arXiv2024

Leveraging Optimization for Adaptive Attacks on Image Watermarks

Untrustworthy users can misuse image generators to synthesize high-quality deepfakes and engage in unethical activities. Watermarking deters misuse by marking generated content with a hidden message, enabling its detection using a secret watermarking key. A core security property of watermarking is robustness, which states that an attacker can only evade detection by substantially degrading image quality. Assessing robustness requires designing an adaptive attack for the specific watermarking algorithm. When evaluating watermarking algorithms and their (adaptive) attacks, it is challenging to determine whether an adaptive attack is optimal, i.e., the best possible attack. We solve this problem by defining an objective function and then approach adaptive attacks as an optimization problem. The core idea of our adaptive attacks is to replicate secret watermarking keys locally by creating surrogate keys that are differentiable and can be used to optimize the attack's parameters. We demonstrate for Stable Diffusion models that such an attacker can break all five surveyed watermarking methods at no visible degradation in image quality. Optimizing our attacks is efficient and requires less than 1 GPU hour to reduce the detection accuracy to 6.3% or less. Our findings emphasize the need for more rigorous robustness testing against adaptive, learnable attackers.

cs.CR↗

arXiv2024

Universal Backdoor Attacks

Web-scraped datasets are vulnerable to data poisoning, which can be used for backdooring deep image classifiers during training. Since training on large datasets is expensive, a model is trained once and re-used many times. Unlike adversarial examples, backdoor attacks often target specific classes rather than any class learned by the model. One might expect that targeting many classes through a naive composition of attacks vastly increases the number of poison samples. We show this is not necessarily true and more efficient, universal data poisoning attacks exist that allow controlling misclassifications from any source class into any target class with a small increase in poison samples. Our idea is to generate triggers with salient characteristics that the model can learn. The triggers we craft exploit a phenomenon we call inter-class poison transferability, where learning a trigger from one class makes the model more vulnerable to learning triggers for other classes. We demonstrate the effectiveness and robustness of our universal backdoor attacks by controlling models with up to 6,000 classes while poisoning only 0.15% of the training dataset. Our source code is available at https://github.com/Ben-Schneider-code/Universal-Backdoor-Attacks.

cs.LG↗

arXiv2023

Identifying and Mitigating the Security Risks of Generative AI

Every major technical invention resurfaces the dual-use dilemma -- the new technology has the potential to be used for good as well as for harm. Generative AI (GenAI) techniques, such as large language models (LLMs) and diffusion models, have shown remarkable capabilities (e.g., in-context learning, code-completion, and text-to-image generation and editing). However, GenAI can be used just as well by attackers to generate new attacks and increase the velocity and efficacy of existing attacks. This paper reports the findings of a workshop held at Google (co-organized by Stanford University and the University of Wisconsin-Madison) on the dual-use dilemma posed by GenAI. This paper is not meant to be comprehensive, but is rather an attempt to synthesize some of the interesting findings from the workshop. We discuss short-term and long-term goals for the community on this topic. We hope this paper provides both a launching point for a discussion on this important topic as well as interesting problems that the research community can work to address.

cs.AI↗

arXiv2023

PTW: Pivotal Tuning Watermarking for Pre-Trained Image Generators

Deepfakes refer to content synthesized using deep generators, which, when misused, have the potential to erode trust in digital media. Synthesizing high-quality deepfakes requires access to large and complex generators only a few entities can train and provide. The threat is malicious users that exploit access to the provided model and generate harmful deepfakes without risking detection. Watermarking makes deepfakes detectable by embedding an identifiable code into the generator that is later extractable from its generated images. We propose Pivotal Tuning Watermarking (PTW), a method for watermarking pre-trained generators (i) three orders of magnitude faster than watermarking from scratch and (ii) without the need for any training data. We improve existing watermarking methods and scale to generators $4 \times$ larger than related work. PTW can embed longer codes than existing methods while better preserving the generator's image quality. We propose rigorous, game-based definitions for robustness and undetectability, and our study reveals that watermarking is not robust against an adaptive white-box attacker who controls the generator's parameters. We propose an adaptive attack that can successfully remove any watermarking with access to only 200 non-watermarked images. Our work challenges the trustworthiness of watermarking for deepfake detection when the parameters of a generator are available. The source code to reproduce our experiments is available at https://github.com/nilslukas/gan-watermark.

cs.LG↗

arXiv2023

Level Up: Private Non-Interactive Decision Tree Evaluation using Levelled Homomorphic Encryption

As machine learning as a service continues gaining popularity, concerns about privacy and intellectual property arise. Users often hesitate to disclose their private information to obtain a service, while service providers aim to protect their proprietary models. Decision trees, a widely used machine learning model, are favoured for their simplicity, interpretability, and ease of training. In this context, Private Decision Tree Evaluation (PDTE) enables a server holding a private decision tree to provide predictions based on a client's private attributes. The protocol is such that the server learns nothing about the client's private attributes. Similarly, the client learns nothing about the server's model besides the prediction and some hyperparameters. In this paper, we propose two novel non-interactive PDTE protocols, XXCMP-PDTE and RCC-PDTE, based on two new non-interactive comparison protocols, XXCMP and RCC. Our evaluation of these comparison operators demonstrates that our proposed constructions can efficiently evaluate high-precision numbers. Specifically, RCC can compare 32-bit numbers in under 10 milliseconds. We assess our proposed PDTE protocols on decision trees trained over UCI datasets and compare our results with existing work in the field. Moreover, we evaluate synthetic decision trees to showcase scalability, revealing that RCC-PDTE can evaluate a decision tree with over 1000 nodes and 16 bits of precision in under 2 seconds. In contrast, the current state-of-the-art requires over 10 seconds to evaluate such a tree with only 11 bits of precision.

cs.CR↗

arXiv2023

Comprehension from Chaos: Towards Informed Consent for Private Computation

Private computation, which includes techniques like multi-party computation and private query execution, holds great promise for enabling organizations to analyze data they and their partners hold while maintaining data subjects' privacy. Despite recent interest in communicating about differential privacy, end users' perspectives on private computation have not previously been studied. To fill this gap, we conducted 22 semi-structured interviews investigating users' understanding of, and expectations for, private computation over data about them. Interviews centered on four concrete data-analysis scenarios (e.g., ad conversion analysis), each with a variant that did not use private computation and another that did (private set intersection, multi-party computation, and privacy preserving query procedures). While participants struggled with abstract definitions of private computation, they found the concrete scenarios enlightening and plausible even though we did not explain the complex cryptographic underpinnings. Private computation increased participants' acceptance of data sharing, but not unconditionally; the purpose of data sharing and analysis was the primary driver of their attitudes. Through collective activities, participants emphasized the importance of detailing the purpose of a computation and clarifying that inputs to private computation are not shared across organizations when describing private computation to end users.

cs.CR↗

arXiv2023

Backdooring Textual Inversion for Concept Censorship

Recent years have witnessed success in AIGC (AI Generated Content). People can make use of a pre-trained diffusion model to generate images of high quality or freely modify existing pictures with only prompts in nature language. More excitingly, the emerging personalization techniques make it feasible to create specific-desired images with only a few images as references. However, this induces severe threats if such advanced techniques are misused by malicious users, such as spreading fake news or defaming individual reputations. Thus, it is necessary to regulate personalization models (i.e., concept censorship) for their development and advancement. In this paper, we focus on the personalization technique dubbed Textual Inversion (TI), which is becoming prevailing for its lightweight nature and excellent performance. TI crafts the word embedding that contains detailed information about a specific object. Users can easily download the word embedding from public websites like Civitai and add it to their own stable diffusion model without fine-tuning for personalization. To achieve the concept censorship of a TI model, we propose leveraging the backdoor technique for good by injecting backdoors into the Textual Inversion embeddings. Briefly, we select some sensitive words as triggers during the training of TI, which will be censored for normal use. In the subsequent generation stage, if the triggers are combined with personalized embeddings as final prompts, the model will output a pre-defined target image rather than images including the desired malicious concept. To demonstrate the effectiveness of our approach, we conduct extensive experiments on Stable Diffusion, a prevailing open-sourced text-to-image model. Our code, data, and results are available at https://concept-censorship.github.io.

cs.CR↗

arXiv2023

Pick your Poison: Undetectability versus Robustness in Data Poisoning Attacks

Deep image classification models trained on vast amounts of web-scraped data are susceptible to data poisoning - a mechanism for backdooring models. A small number of poisoned samples seen during training can severely undermine a model's integrity during inference. Existing work considers an effective defense as one that either (i) restores a model's integrity through repair or (ii) detects an attack. We argue that this approach overlooks a crucial trade-off: Attackers can increase robustness at the expense of detectability (over-poisoning) or decrease detectability at the cost of robustness (under-poisoning). In practice, attacks should remain both undetectable and robust. Detectable but robust attacks draw human attention and rigorous model evaluation or cause the model to be re-trained or discarded. In contrast, attacks that are undetectable but lack robustness can be repaired with minimal impact on model accuracy. Our research points to intrinsic flaws in current attack evaluation methods and raises the bar for all data poisoning attackers who must delicately balance this trade-off to remain robust and undetectable. To demonstrate the existence of more potent defenders, we propose defenses designed to (i) detect or (ii) repair poisoned models using a limited amount of trusted image-label pairs. Our results show that an attacker who needs to be robust and undetectable is substantially less threatening. Our defenses mitigate all tested attacks with a maximum accuracy decline of 2% using only 1% of clean data on CIFAR-10 and 2.5% on ImageNet. We demonstrate the scalability of our defenses by evaluating large vision-language models, such as CLIP. Attackers who can manipulate the model's parameters pose an elevated risk as they can achieve higher robustness at low detectability compared to data poisoning attackers.

cs.CR↗

arXiv2023

Investigating Membership Inference Attacks under Data Dependencies

Training machine learning models on privacy-sensitive data has become a popular practice, driving innovation in ever-expanding fields. This has opened the door to new attacks that can have serious privacy implications. One such attack, the Membership Inference Attack (MIA), exposes whether or not a particular data point was used to train a model. A growing body of literature uses Differentially Private (DP) training algorithms as a defence against such attacks. However, these works evaluate the defence under the restrictive assumption that all members of the training set, as well as non-members, are independent and identically distributed. This assumption does not hold for many real-world use cases in the literature. Motivated by this, we evaluate membership inference with statistical dependencies among samples and explain why DP does not provide meaningful protection (the privacy parameter $ε$ scales with the training set size $n$) in this more general case. We conduct a series of empirical evaluations with off-the-shelf MIAs using training sets built from real-world data showing different types of dependencies among samples. Our results reveal that training set dependencies can severely increase the performance of MIAs, and therefore assuming that data samples are statistically independent can significantly underestimate the performance of MIAs.

cs.CR↗

arXiv2022

Towards Robust Dataset Learning

Adversarial training has been actively studied in recent computer vision research to improve the robustness of models. However, due to the huge computational cost of generating adversarial samples, adversarial training methods are often slow. In this paper, we study the problem of learning a robust dataset such that any classifier naturally trained on the dataset is adversarially robust. Such a dataset benefits the downstream tasks as natural training is much faster than adversarial training, and demonstrates that the desired property of robustness is transferable between models and data. In this work, we propose a principled, tri-level optimization to formulate the robust dataset learning problem. We show that, under an abstraction model that characterizes robust vs. non-robust features, the proposed method provably learns a robust dataset. Extensive experiments on MNIST, CIFAR10, and TinyImageNet demostrate the effectiveness of our algorithm with different network initializations and architectures.

cs.CV↗

arXiv2022

Faster Secure Comparisons with Offline Phase for Efficient Private Set Intersection

In a Private section intersection (PSI) protocol, Alice and Bob compute the intersection of their respective sets without disclosing any element not in the intersection. PSI protocols have been extensively studied in the literature and are deployed in industry. With state-of-the-art protocols achieving optimal asymptotic complexity, performance improvements are rare and can only improve complexity constants. In this paper, we present a new private, extremely efficient comparison protocol that leads to a PSI protocol with low constants. A useful property of our comparison protocol is that it can be divided into an online and an offline phase. All expensive cryptographic operations are performed during the offline phase, and the online phase performs only four fast field operations per comparison. This leads to an incredibly fast online phase, and our evaluation shows that it outperforms related work, including KKRT (CCS 16), VOLE-PSI (EuroCrypt 21), and OKVS (Crypto 21). We also evaluate standard approaches to implement the offline phase using different trust assumptions: cryptographic, hardware, and a third party (dealer model).

cs.CR↗

arXiv2022

Selective MPC: Distributed Computation of Differentially Private Key-Value Statistics

Key-value data is a naturally occurring data type that has not been thoroughly investigated in the local trust model. Existing local differentially private (LDP) solutions for computing statistics over key-value data suffer from the inherent accuracy limitations of each user adding their own noise. Multi-party computation (MPC) maintains better accuracy than LDP and similarly does not require a trusted central party. However, naively applying MPC to key-value data results in prohibitively expensive computation costs. In this work, we present selective multi-party computation, a novel approach to distributed computation that leverages DP leakage to efficiently and accurately compute statistics over key-value data. By providing each party with a view of a random subset of the data, we can capture subtractive noise. We prove that our protocol satisfies pure DP and is provably secure in the combined DP/MPC model. Our empirical evaluation demonstrates that we can compute statistics over 10,000 keys in 20 seconds and can scale up to 30 servers while obtaining results for a single key in under a second.

cs.CR↗

arXiv2022

IHOP: Improved Statistical Query Recovery against Searchable Symmetric Encryption through Quadratic Optimization

Effective query recovery attacks against Searchable Symmetric Encryption (SSE) schemes typically rely on auxiliary ground-truth information about the queries or dataset. Query recovery is also possible under the weaker statistical auxiliary information assumption, although statistical-based attacks achieve lower accuracy and are not considered a serious threat. In this work we present IHOP, a statistical-based query recovery attack that formulates query recovery as a quadratic optimization problem and reaches a solution by iterating over linear assignment problems. We perform an extensive evaluation with five real datasets, and show that IHOP outperforms all other statistical-based query recovery attacks under different parameter and leakage configurations, including the case where the client uses some access-pattern obfuscation defenses. In some cases, our attack achieves almost perfect query recovery accuracy. Finally, we use IHOP in a frequency-only leakage setting where the client's queries are correlated, and show that our attack can exploit query dependencies even when PANCAKE, a recent frequency-hiding defense by Grubbs et al., is applied. Our findings indicate that statistical query recovery attacks pose a severe threat to privacy-preserving SSE schemes.

cs.CR↗

arXiv2022

The Limits of Word Level Differential Privacy

As the issues of privacy and trust are receiving increasing attention within the research community, various attempts have been made to anonymize textual data. A significant subset of these approaches incorporate differentially private mechanisms to perturb word embeddings, thus replacing individual words in a sentence. While these methods represent very important contributions, have various advantages over other techniques and do show anonymization capabilities, they have several shortcomings. In this paper, we investigate these weaknesses and demonstrate significant mathematical constraints diminishing the theoretical privacy guarantee as well as major practical shortcomings with regard to the protection against deanonymization attacks, the preservation of content of the original sentences as well as the quality of the language output. Finally, we propose a new method for text anonymization based on transformer based language models fine-tuned for paraphrasing that circumvents most of the identified weaknesses and also offers a formal privacy guarantee. We evaluate the performance of our method via thorough experimentation and demonstrate superior performance over the discussed mechanisms.

cs.CR↗