During the past decade, Deep Neural Networks (DNNs) proved their value on a large variety of subjects. However despite their high value and public accessibility, the protection of the intellectual property of DNNs is still an issue and an emerging research field. Recent works have successfully extracted fully-connected DNNs using cryptanalytic methods in hard-label settings, proving that it was possible to copy a DNN with high fidelity, i.e., high similitude in the output predictions....

As privacy concerns have arisen in machine learning, privacy-preserving machine learning (PPML) has received significant attention. Fully homomorphic encryption (FHE) and secure multi-party computation (MPC) are representative building blocks for PPML. However, in PPML protocols based on FHE and MPC, interaction between the client (who provides encrypted input data) and the evaluator (who performs the computation) is essential to obtain the final result in plaintext. Functional encryption...

We present a novel approach for training neural networks that leverages packed Shamir secret sharing scheme. For specific training protocols based on Shamir scheme, we demonstrate how to realize the conversion between packed sharing and Shamir sharing without additional communication overhead. We begin by introducing a method to locally convert between Shamir sharings with secrets stored at different slots. Building upon this conversion, we achieve free conversion from packed sharing to...

Fully Homomorphic Encryption (FHE) is a promising privacy-enhancing technique that enables secure and private data processing on untrusted servers, such as privacy-preserving neural network (NN) evaluations. However, its practical application presents significant challenges. Limitations in how data is stored within homomorphic ciphertexts and restrictions on the types of operations that can be performed create computational bottlenecks. As a result, a growing body of research focuses on...

Deep neural networks (DNNs) are valuable assets, yet their public accessibility raises security concerns about parameter extraction by malicious actors. Recent work by Carlini et al. (Crypto’20) and Canales- Martínez et al. (Eurocrypt’24) has drawn parallels between this issue and block cipher key extraction via chosen plaintext attacks. Leveraging differential cryptanalysis, they demonstrated that all the weights and biases of black-box ReLU-based DNNs could be inferred using a polynomial...

Recently, deep learning-based side-channel analysis (DLSCA) has emerged as a serious threat against cryptographic implementations. These methods can efficiently break implementations protected with various countermeasures while needing limited manual intervention. To effectively protect implementation, it is therefore crucial to be able to interpret \textbf{how} these models are defeating countermeasures. Several works have attempted to gain a better understanding of the mechanics of these...

The machine learning problem of extracting neural network parameters has been proposed for nearly three decades. Functionally equivalent extraction is a crucial goal for research on this problem. When the adversary has access to the raw output of neural networks, various attacks, including those presented at CRYPTO 2020 and EUROCRYPT 2024, have successfully achieved this goal. However, this goal is not achieved when neural networks operate under a hard-label setting where the raw output...

Privacy-Preserving Machine Learning is one of the most relevant use cases for Secure Multiparty Computation (MPC). While private training of large neural networks such as VGG-16 or ResNet-50 on state-of-the-art datasets such as Imagenet is still out of reach, given the performance overhead of MPC, private inference is starting to achieve practical runtimes. However, we show that in contrast to plaintext machine learning, the usage of GPU acceleration for both linear and nonlinear neural...

Privacy-preserving machine learning (PPML) enables multiple distrusting parties to jointly train ML models on their private data without revealing any information beyond the final trained models. In this work, we study the client-aided two-server setting where two non-colluding servers jointly train an ML model on the data held by a large number of clients. By involving the clients in the training process, we develop efficient protocols for training algorithms including linear regression,...

Recent advancements in transformers have revolutionized machine learning, forming the core of Large language models (LLMs). However, integrating these systems into everyday applications raises privacy concerns as client queries are exposed to model owners. Secure multiparty computation (MPC) allows parties to evaluate machine learning applications while keeping sensitive user inputs and proprietary models private. Due to inherent MPC costs, recent works introduce model-specific optimizations...

With concerns about data privacy growing in a connected world, cryptography researchers have focused on fully homomorphic encryption (FHE) for promising machine learning as a service solutions. Recent advancements have lowered the computational cost by several orders of magnitude, but the latency of fully homomorphic neural networks remains a barrier to adoption. This work proposes using multi-exit neural networks (MENNs) to accelerate the FHE inference. MENNs are network architectures that...

We present a variant of Function Secret Sharing (FSS) schemes tailored for point, comparison, and interval functions, featuring compact key sizes at the expense of additional comparison. While existing FSS constructions are primarily geared towards $2$-party scenarios, exceptions such as the work by Boyle et al. (Eurocrypt 2015) and Riposte (S&P 2015) have introduced FSS schemes for $p$-party scenarios ($p \geq 3$). This paper aims to achieve the most compact $p$-party FSS key size to date....

Neural network inference as a service enables a cloud server to provide inference services to clients. To ensure the privacy of both the cloud server's model and the client's data, secure neural network inference is essential. Binarized neural networks (BNNs), which use binary weights and activations, are often employed to accelerate inference. However, achieving secure BNN inference with secure multi-party computation (MPC) is challenging because MPC protocols cannot directly operate on...

In the field of Artificial Intelligence (AI), convolution operations have primarily been used in Convolutional Neural Networks (CNNs). However, its utility is increasing with the appearance of convolution integrated transformers or state space models where convolution is a constituent element. In the field of private AI, generalized algorithm, multiplexed parallel convolution was recently proposed to implement CNNs based on the Homomorphic Encryption scheme, residue number system variant...

In recent years, cloud vendors have started to supply paid services for data analysis by providing interfaces of their well-trained neural network models. However, customers lack tools to verify whether outcomes supplied by cloud vendors are correct inferences from particular models, in the face of lazy or malicious vendors. The cryptographic primitive called zero-knowledge proof (ZKP) addresses this problem. It enables the outcomes to be verifiable without leaking information about the...

Transformer neural networks have gained significant traction since their introduction, becoming pivotal across diverse domains. Particularly in large language models like Claude and ChatGPT, the transformer architecture has demonstrated remarkable efficacy. This paper provides a concise overview of transformer neural networks and delves into their security considerations, focusing on covert channel attacks and their implications for the safety of large language models. We present a covert...

In this paper we address the use of Neural Networks (NN) for the assessment of the quality and hence safety of several Random Number Generators (RNGs), focusing both on the vulnerability of classical Pseudo Random Number Generators (PRNGs), such as Linear Congruential Generators (LCGs) and the RC4 algorithm, and extending our analysis to non-conventional data sources, such as Quantum Random Number Generators (QRNGs) based on Vertical-Cavity Surface- Emitting Laser (VCSEL). Among the...

Classifying images has become a straightforward and accessible task, thanks to the advent of Deep Neural Networks. Nevertheless, not much attention is given to the privacy concerns associated with sensitive data contained in images. In this study, we propose a solution to this issue by exploring an intersection between Machine Learning and cryptography. In particular, Fully Homomorphic Encryption (FHE) emerges as a promising solution, as it enables computations to be performed on encrypted...

At CRYPTO 2019, Gohr demonstrated that differential-neural distinguishers (DNDs) for Speck32/64 can learn more features than classical cryptanalysis's differential distribution tables (DDT). Furthermore, a non-classical key recovery procedure is devised by combining the Upper Confidence Bound (UCB) strategy and the BayesianKeySearch algorithm. Consequently, the time complexity of 11-round key recovery attacks on Speck32/64 is significantly reduced compared with the state-of-the-art results...

This research extends Abadi and Andersen's exploration of neural networks using secret keys for information protection in multiagent systems. Focusing on enhancing confidentiality properties, we employ end-to-end adversarial training with neural networks Alice, Bob, and Eve. Unlike prior work limited to 64-bit messages, our study spans message sizes from 4 to 1024 bits, varying batch sizes and training steps. An innovative aspect involves training model Bob to approach a minimal error value...

In side-channel analysis (SCA), the success of an attack is largely dependent on the dataset sizes and the number of instances in each class. The generation of synthetic traces can help to improve attacks like profiling attacks. However, manually creating synthetic traces from actual traces is arduous. Therefore, automating this process of creating artificial traces is much needed. Recently, diffusion models have gained much recognition after beating another generative model known as...

A zero-knowledge proof of training (zkPoT) enables a party to prove that they have correctly trained a committed model based on a committed dataset without revealing any additional information about the model or the dataset. An ideal zkPoT should offer provable security and privacy guarantees, succinct proof size and verifier runtime, and practical prover efficiency. In this work, we present \name, a zkPoT targeted for deep neural networks (DNNs) that achieves all these goals at once. Our...

The past years have witnessed a considerable increase in research efforts put into neural network-assisted profiled side-channel analysis (SCA). Studies have also identified challenges, e.g., closing the gap between metrics for machine learning (ML) classification and side-channel attack evaluation. In fact, in the context of NN-assisted SCA, the NN’s output distribution forms the basis for successful key recovery. In this respect, related work has covered various aspects of integrating...

Side-channel analysis based on machine learning, especially neural networks, has gained significant attention in recent years. However, many existing methods still suffer from certain limitations. Despite the inherent capability of neural networks to extract features, there remains a risk of extracting irrelevant information. The heavy reliance on profiled traces makes it challenging to adapt to remote attack scenarios with limited profiled traces. Besides, attack traces also contain...

In recent years, deep learning-based side-channel analysis (DLSCA) has become an active research topic within the side-channel analysis community. The well-known challenge of hyperparameter tuning in DLSCA encouraged the community to use methods that reduce the effort required to identify an optimal model. One of the successful methods is ensemble learning. While ensemble methods have demonstrated their effectiveness in DLSCA, particularly with AES-based datasets, their efficacy in analyzing...

In this work, we introduce FANNG-MPC, a versatile secure multi-party computation framework capable to offer active security for privacy preserving machine learning as a service (MLaaS). Derived from the now deprecated SCALE-MAMBA, FANNG is a data-oriented fork, featuring novel set of libraries and instructions for realizing private neural networks, effectively reviving the popular framework. To the best of our knowledge, FANNG is the first MPC framework to offer actively secure MLaaS in the...

Private computation of nonlinear functions, such as Rectified Linear Units (ReLUs) and max-pooling operations, in deep neural networks (DNNs) poses significant challenges in terms of storage, bandwidth, and time consumption. To address these challenges, there has been a growing interest in utilizing privacy-preserving techniques that leverage polynomial activation functions and kernelized convolutions as alternatives to traditional ReLUs. However, these alternative approaches often suffer...

The privacy-preserving machine learning (PPML) has gained growing importance over the last few years. One of the biggest challenges is to improve the efficiency of PPML so that the communication and computation costs of PPML are affordable for large machine learning models such as deep learning. As we know, linear algebra such as matrix multiplication occupies a significant part of the computation in deep learning such as deep convolutional neural networks (CNN). Thus, it is desirable to...

Secure Multiparty Computation (MPC) protocols enable secure evaluation of a circuit by several parties, even in the presence of an adversary who maliciously corrupts all but one of the parties. These MPC protocols are constructed using the well-known secret-sharing-based paradigm (SPDZ and SPD$\mathbb{Z}_{2^k}$), where the protocols ensure security against a malicious adversary by computing Message Authentication Code (MAC) tags on the input shares and then evaluating the circuit with these...

The escalation of hazards to safety and hijacking of digital networks are among the strongest perilous difficulties that must be addressed in the present day. Numerous safety procedures were set up to track and recognize any illicit activity on the network's infrastructure. IDS are the best way to resist and recognize intrusions on internet connections and digital technologies. To classify network traffic as normal or anomalous, Machine Learning (ML) classifiers are increasingly utilized. An...

Billions of dollars and countless GPU hours are currently spent on training Deep Neural Networks (DNNs) for a variety of tasks. Thus, it is essential to determine the difficulty of extracting all the parameters of such neural networks when given access to their black-box implementations. Many versions of this problem have been studied over the last 30 years, and the best current attack on ReLU-based deep neural networks was presented at Crypto’20 by Carlini, Jagielski, and Mironov. It...

Homomorphic Encryption (HE) presents a promising solution to securing neural networks for Machine Learning as a Service (MLaaS). Despite its potential, the real-time applicability of current HE-based solutions remains a challenge, and the diversity in network structures often results in inefficient implementations and maintenance. To address these issues, we introduce a unified and compact network structure for real-time inference in convolutional neural networks based on HE. We further...

In CRYPTO 2019, Gohr showed that well-trained neural networks could perform cryptanalytic distinguishing tasks superior to differential distribution table (DDT)-based distinguishers. This suggests that the differential-neural distinguisher (ND) may use additional information besides pure ciphertext differences. However, the explicit knowledge beyond differential distribution is still unclear. In this work, we provide explicit rules that can be used alongside DDTs to enhance the effectiveness...

Cryptographic proof systems provide integrity, fairness, and privacy in applications that outsource data processing tasks. However, general-purpose proof systems do not scale well to large inputs. At the same time, ad-hoc solutions for concrete applications - e.g., machine learning or image processing - are more efficient but lack modularity, hence they are hard to extend or to compose with other tools of a data-processing pipeline. In this paper, we combine the performance of tailored...

Data security has become a paramount concern in the age of data driven applications, necessitating the deployment of robust encryption techniques. This paper presents an in-depth investigation into the strength and randomness of the keystream generated by the Grain cipher, a widely employed stream cipher in secure communication systems. To achieve this objective, we propose the construction of sophisticated deep learning models for keystream prediction and evaluation. The implications of...

Side-channel analysis certification is a process designed to certify the resilience of cryptographic hardware and software implementations against side-channel attacks. In certain cases, third-party evaluations by external companies or departments are necessary due to limited budget, time, or even expertise with the penalty of a significant exchange of sensitive information during the evaluation process. In this work, we investigate the potential of Homomorphic Encryption (HE) in...

Side-channel attacks against cryptographic implementations are mitigated by the application of masking and hiding countermeasures. Hiding countermeasures attempt to reduce the Signal-to-Noise Ratio of measurements by adding noise or desynchronization effects during the execution of the cryptographic operations. To bypass these protections, attackers adopt signal processing techniques such as pattern alignment, filtering, averaging, or resampling. Convolutional neural networks have shown the...

The recent advancements in deep learning have brought about significant changes in various aspects of people's lives. Meanwhile, these rapid developments have raised concerns about the legitimacy of the training process of deep neural networks. To protect the intellectual properties of AI developers, directly examining the training process by accessing the model parameters and training data is often prohibited for verifiers. In response to this challenge, we present zero-knowledge deep...

Convolutional neural networks (CNNs) offer unrivaled performance in profiling side-channel analysis. This claim is corroborated by numerous results where CNNs break targets protected with masking and hiding countermeasures. One hiding countermeasure is commonly investigated in related works - desynchronization (misalignment). The conclusions usually state that CNNs can break desynchronization as they are shift-invariant. This paper investigates that claim in more detail and reveals that the...

Deep neural networks (DNNs) represent a powerful technique for assessing cryptographic security concerning side-channel analysis (SCA) due to their ability to aggregate leakages automatically, rendering attacks more efficient without preprocessing. Nevertheless, despite their effectiveness, DNNs employed in SCA are predominantly black-box algorithms, posing considerable interpretability challenges. In this paper, we propose a novel technique called Key Guessing Occlusion (KGO) that...

We introduce a deep learning ensemble (NNBits) as a tool for bit-profiling and evaluation of cryptographic (pseudo) random bit sequences. Onthe one hand, we show how to use NNBits ensemble to ex-plain parts of the seminal work of Gohr [16]: Gohr’s depth-1 neural distinguisher reaches a test accuracy of 78.3% in round 6 for SPECK32/64 [3]. Using the bit-level information provided by NNBits we can partially ex- plain the accuracy obtained by Gohr (78.1% vs. 78.3%). This is achieved by...

AI-as-a-Service has emerged as an important trend for supporting the growth of the digital economy. Digital service providers make use of their vast amount of user data to train AI models (such as image recognitions, financial modelling and pandemic modelling etc.) and offer them as a service on the cloud. While there are convincing advantages for using such third-party models, the fact that users need to upload their data to the cloud is bound to raise serious privacy concerns,...

Secure Machine Learning as a Service is a viable solution where clients seek secure delegation of the ML computation while protecting their sensitive data. We propose an efficient method to securely evaluate deep standard convolutional neural networks based on CKKS fully homomorphic encryption, in the manner of batch inference. In this paper, we introduce a packing method called Channel-by-Channel Packing that maximizes the slot compactness and single-instruction-multipledata capabilities in...

In today’s interconnected world, data has become a valuable asset, leading to a growing interest in protecting it through techniques such as privacy-preserving computation. Two well-known approaches are multi-party computation and homomorphic encryption with use cases such as privacy-preserving machine learning evaluating or training neural networks. For multi-party computation, one of the fundamental arithmetic operations is the secure multiplication in the malicious security model and by...

Spiking neural networks gain attention due to low power properties and event-based operation, making them suitable for usage in resource constrained embedded devices. Such edge devices allow physical access opening the door for side-channel analysis. In this work, we reverse engineer the parameters of a feed-forward spiking neural network implementation with correlation power analysis. Localized measurements of electro-magnetic emanations enable our attack, despite inherent parallelism and...

Homomorphic encryption (HE) enables calculating on encrypted data, which makes it possible to perform privacy- preserving neural network inference. One disadvantage of this technique is that it is several orders of magnitudes slower than calculation on unencrypted data. Neural networks are commonly trained using floating-point, while most homomorphic encryption libraries calculate on integers, thus requiring a quantisation of the neural network. A straightforward approach would be to...

We conduct a systematic study of solving the learning parity with noise problem (LPN) using neural networks. Our main contribution is designing families of two-layer neural networks that practically outperform classical algorithms in high-noise, low-dimension regimes. We consider three settings where the numbers of LPN samples are abundant, very limited, and in between. In each setting we provide neural network models that solve LPN as fast as possible. For some settings we are also able to...

During the last decade, there has been a stunning progress in the domain of AI with adoption in both safety-critical and security-critical applications. A key requirement for this is highly trained Machine Learning (ML) models, which are valuable Intellectual Property (IP) of the respective organizations. Naturally, these models have become targets for model recovery attacks through side-channel leakage. However, majority of the attacks reported in literature are either on simple embedded...

Machine learning (ML) has seen a strong rise in popularity in recent years and has become an essential tool for research and industrial applications. Given the large amount of high quality data needed and the often sensitive nature of ML data, privacy-preserving collaborative ML is of increasing importance. In this paper, we introduce new actively secure multiparty computation (MPC) protocols which are specially optimized for privacy-preserving machine learning applications. We concentrate...

Fully homomorphic encryption (FHE) is an encryption method that allows to perform computation on encrypted data, without decryption. FHE preserves the privacy of the users of online services that handle sensitive data, such as health data, biometrics, credit scores and other personal information. A common way to provide a valuable service on such data is through machine learning and, at this time, Neural Networks are the dominant machine learning model for unstructured data. In this work...

Secure inference of deep convolutional neural networks (CNNs) under RNS-CKKS involves polynomial approximation of unsupported non-linear activation functions. However, existing approaches have three main limitations: 1) Inflexibility: The polynomial approximation and associated homomorphic evaluation architecture are customized manually for each CNN architecture and do not generalize to other networks. 2) Suboptimal Approximation: Each activation function is approximated instead of the...

Secure neural network inference has been a promising solution to private Deep-Learning-as-a-Service, which enables the service provider and user to execute neural network inference without revealing their private inputs. However, the expensive overhead of current schemes is still an obstacle when applied in real applications. In this work, we present \textsc{Meteor}, an online communication-efficient and fast secure 3-party computation neural network inference system aginst semi-honest...

The usage of convolutional neural networks (CNNs) to break cryptographic systems through hardware side-channels has enabled fast and adaptable attacks on devices like smart cards and TPMs. Current literature proposes fixed CNN architectures designed by domain experts to break such systems, which is time-consuming and unsuitable for attacking a new system. Recently, an approach using neural architecture search (NAS), which is able to acquire a suitable architecture automatically, has been...

Masking has become one of the most effective approaches for securing hardware designs against side-channel attacks. Regardless of the effort put into correctly implementing masking schemes on a field-programmable gate array (FPGA), leakage can be unexpectedly observed. This is due to the fact that the assumption underlying all masked designs, i.e., the leakages of different shares are independent of each other, may no longer hold in practice. In this regard, extreme temperatures have been...

Previous studies on fault analysis have demonstrated promising potential in compromising cryptographic security. However, these fault analysis methods are limited in practical impact due to methodological constraints and the substantial requirement of faulty information such as correct and faulty ciphertexts. Additionally, while deep learning techniques have been widely applied to side-channel analysis (SCA) in recent years and have shown superior performance compared with traditional...

Deep learning is very efficient at breaking masked implementations even when the attacker does not assume knowledge of the masks. However, recent works pointed out a significant challenge: overcoming the initial learning plateau. This paper discusses the advantages of multi-task learning to break through the initial plateau consistently. We investigate different ways of applying multi-task learning against masked AES implementations (via the ASCAD-r, ASCAD-v2, and CHESCTF-2023 datasets)...

With the recent development of quantum computers, various studies on quantum artificial intelligence technology are being conducted. Quantum artificial intelligence can improve performance in terms of accuracy and memory usage compared to deep learning on classical computers. In this work, we proposed an attack technique that recovers keys by learning patterns in cryptographic algorithms by applying quantum artificial intelligence to cryptanalysis. Cryptanalysis was performed in the current...

In CRYPTO 2019, Gohr opens up a new direction for cryptanalysis. He successfully applied deep learning to differential cryptanalysis against the NSA block cipher SPECK32/64, achieving higher accuracy than traditional differential distinguishers. Until now, one of the mainstream research directions is increasing the training sample size and utilizing different neural networks to improve the accuracy of neural distinguishers. This conversion mindset may lead to a huge number of parameters,...

Logic locking/obfuscation secures hardware designs from untrusted entities throughout the globalized semiconductor supply chain. Machine learning (ML) recently challenged the security of locking: such attacks successfully captured the locking-induced, structural design modifications to decipher obfuscated gates. Although routing obfuscation eliminates this threat, more recent attacks exposed new vulnerabilities, like link formation, breaking such schemes. Thus, there is still a need for...

CRYSTALS-Kyber has been selected by the NIST as a public-key encryption and key encapsulation mechanism to be standardized. It is also included in the NSA's suite of cryptographic algorithms recommended for national security systems. This makes it important to evaluate the resistance of CRYSTALS-Kyber's implementations to side-channel attacks. The unprotected and first-order masked software implementations have been already analysed. In this paper, we present deep learning-based message...

Differential cryptanalysis is a block cipher analysis technology that infers a key by using the difference characteristics. Input differences can be distinguished using a good difference characteristic, and this distinguishing task can lead to key recovery. Artificial neural networks are a good solution for distinguishing tasks. For this reason, recently, neural distinguishers have been actively studied. We propose a distinguisher based on a quantum-classical hybrid neural network by...

In the seminal work published by Gohr in CRYPTO 2019, neural networks were successfully exploited to perform differential attacks on Speck32/64, the smallest member in the block cipher family Speck. The deep learning aided key-recovery attack by Gohr achieves considerable improvement in terms of time complexity upon the state-of-the-art result from the conventional cryptanalysis method. A further question is whether the advantage of deep learning aided attacks can be kept on large-state...

Data privacy concerns are increasing significantly in the context of Internet of Things, cloud services, edge computing, artificial intelligence applications, and other applications enabled by next generation networks. Homomorphic Encryption addresses privacy challenges by enabling multiple operations to be performed on encrypted messages without decryption. This paper comprehensively addresses homomorphic encryption from both theoretical and practical perspectives. The paper delves into the...

Neural networks (NNs) have become one of the most important tools for artificial intelligence (AI). Well-designed and trained NNs can perform inference (e.g., make decisions or predictions) on unseen inputs with high accuracy. Using NNs often involves sensitive data: depending on the specific use case, the input to the NN and/or the internals of the NN (e.g., the weights and biases) may be sensitive. Thus, there is a need for techniques for performing NN inference securely, ensuring that...

Deep neural networks (DNN) have become a significant threat to the security of cryptographic implementations with regards to side-channel analysis (SCA), as they automatically combine the leakages without any preprocessing needed, leading to a more efficient attack. However, these DNNs for SCA remain mostly black-box algorithms that are very difficult to interpret. Benamira \textit{et al.} recently proposed an interpretable neural network called Truth Table Deep Convolutional Neural Network...

Secure multiparty computation (MPC) is increasingly being used to address privacy issues in various applications. The recent work of Alon et al. (CRYPTO'20) identified the shortcomings of traditional MPC and defined a Friends-and-Foes (FaF) security notion to address the same. We showcase the need for FaF security in real-world applications such as dark pools. This subsequently necessitates designing concretely efficient FaF-secure protocols. Towards this, keeping efficiency at the center...

Deep neural networks have proven effective for second-order profiling side-channel attacks, even in a black-box setting with no prior knowledge of masks and implementation details. While such attacks have been successful, no explanations were provided for understanding why a variety of deep neural networks can (or cannot) learn high-order leakages and what the limitations are. In other words, we lack the explainability on neural network layers combining (or not) unknown and random secret...

The deep learning-based side-channel analysis gave some of the most prominent side-channel attacks against protected targets in the past few years. To this end, the research community's focus has been on creating 1) powerful and 2) (if possible) minimal multilayer perceptron or convolutional neural network architectures. Currently, we see that computationally intensive hyperparameter tuning methods (e.g., Bayesian optimization or reinforcement learning) provide the best results. However,...

We implement training of neural networks in secure multi-party computation (MPC) using quantization commonly used in said setting. We are the first to present an MNIST classifier purely trained in MPC that comes within 0.2 percent of the accuracy of the same convolutional neural network trained via plaintext computation. More concretely, we have trained a network with two convolutional and two dense layers to 99.2% accuracy in 3.5 hours (under one hour for 99% accuracy). We have also...

Public-key cryptography, including conventional cryptosystems and post-quantum cryptography, involves computation-intensive workloads. With noticing the extraordinary computing power of AI accelerators, in this paper, we further explore the feasibility to introduce AI accelerators into high-performance cryptographic computing. Since AI accelerators are dedicated to machine learning or neural networks, the biggest challenge is how to transform cryptographic workloads into their operations,...

Creating a good deep learning (DL) model is an art which requires expertise in DL and a large set of labeled data for training neural networks. Neither is readily available. In this paper, we introduce a method which enables us to achieve good results with bad DL models. We use simple multilayer perceptron (MLP) networks, trained on a small dataset, which make strongly biased predictions if used without the proposed method. The core idea is to extend the attack dataset so that at least one...

Secure machine learning (ML) inference can provide meaningful privacy guarantees to both the client (holding sensitive input) and the server (holding sensitive weights of the ML model) while realizing inference-as-a-service. Although many specialized protocols exist for this task, including those in the preprocessing model (where a majority of the overheads are moved to an input independent offline phase), they all still suffer from large online complexity. Specifically, the protocol phase...

The growing volumes of data being collected and its analysis to provide better services are creating worries about digital privacy. To address privacy concerns and give practical solutions, the literature has relied on secure multiparty computation. However, recent research has mostly focused on the small-party honest-majority setting of up to four parties, noting efficiency concerns. In this work, we extend the strategies to support a larger number of participants in an honest-majority...

At CRYPTO'19, A. Gohr proposed neural distinguishers for the lightweight block cipher Speck32/64, achieving better results than the state-of-the-art at that point. However, the motivation for using that particular architecture was not very clear, leading us to investigate whether a smaller and/or better performing neural distinguisher exists. This paper studies the depth-10 and depth-1 neural distinguishers proposed by Gohr with the aim of finding out whether smaller or better-performing...

Profiling side-channel analysis allows evaluators to estimate the worst-case security of a target. When security evaluations relax the assumptions about the adversary's knowledge, profiling models may easily be sub-optimal due to the inability to extract the most informative points of interest from the side-channel measurements. When used for profiling attacks, deep neural networks can learn strong models without feature selection with the drawback of expensive hyperparameter tuning....

We build a library SecFloat for secure 2-party computation (2PC) of 32-bit single-precision floating-point operations and math functions. The existing functionalities used in cryptographic works are imprecise and the precise functionalities used in standard libraries are not crypto-friendly, i.e., they use operations that are cheap on CPUs but have exorbitant cost in 2PC. SecFloat bridges this gap with its novel crypto-friendly precise functionalities. Compared to the prior cryptographic...

Secure two-party neural network inference (2PC-NN) can offer privacy protection for both the client and the server and is a promising technique in the machine-learning-as-a-service setting. However, the large overhead of the current 2PC-NN in- ference systems is still being a headache, especially when applied to deep neural networks such as ResNet50. In this work, we present Cheetah, a new 2PC-NN inference system that is faster and more communication-efficient than state-of-the-arts. The...

Comparisons or Inequality Tests are an essential building block of Rectified Linear Unit functions (ReLU's), ever more present in Machine Learning, specifically in Neural Networks. Motivated by the increasing interest in privacy-preserving Artificial Intelligence, we explore the current state of the art of privacy preserving comparisons over Multi-Party Computation (MPC). We then introduce constant round variations and combinations, which are compatible with customary fixed point arithmetic...

In CRYPTO'19, Gohr introduced a novel cryptanalysis method by developing a differential-neural distinguisher using neural networks as the underlying distinguisher. He effectively integrated this distinguisher with classical differentials, facilitating a 12-round key recovery attack on Speck32/64 (from a total of 22 rounds). Bao et al. refined the concept of neutral bits, enabling key recovery attacks up to 13 rounds for Speck32/64 and 16 rounds (from a total of 32) for Simon32/64. Our...

The topic of this contribution is the cryptanalytic use of spurious keys, i.e. non-target keys returned by exhaustive key search. We show that the counting of spurious keys allows the construction of distinguishing attacks against block ciphers that are generically expected to start working at (marginally) lower computational cost than is required to find the target key by exhaustive search. We further show that if a brute force distinguisher does return a strong distinguishing signal,...

Recently, the standard ResNet-20 network was successfully implemented on residue number system variant Cheon-Kim-Kim-Song (RNS-CKKS) scheme using bootstrapping, but the implementation lacks practicality due to high latency and low security level. To improve the performance, we first minimize total bootstrapping runtime using multiplexed parallel convolution that collects sparse output data for multiple channels compactly. We also propose the \emph{imaginary-removing bootstrapping} to prevent...

The adoption of deep neural networks for profiling side-channel attacks (SCA) opened new perspectives for leakage detection. Recent publications showed that cryptographic implementations featuring different countermeasures could be broken without feature selection or trace preprocessing. This success comes with a high price: extensive hyperparameter search to find optimal deep learning models. As deep learning models usually suffer from overfitting due to their high fitting capacity, it is...

The use of deep learning techniques to perform side-channel analysis attracted the attention of many researchers as they obtained good performances with them. Unfortunately, the understanding of the neural networks used to perform side-channel attacks is not very advanced yet. In this paper, we propose to contribute to this direction by studying the impact of some particular deep learning techniques for tackling side-channel attack problems. More precisely, we propose to focus on three...

Intellectual Property (IP) thefts of trained machine learning (ML) models through side-channel attacks on inference engines are becoming a major threat. Indeed, several recent works have shown reverse engineering of the model internals using such attacks, but the research on building defenses is largely unexplored. There is a critical need to efficiently and securely transform those defenses from cryptography such as masking to ML frameworks. Existing works, however, revealed that a...

This paper introduces Otti, a general-purpose compiler for (zk)SNARKs that provides support for numerical optimization problems. Otti produces efficient arithmetizations of programs that contain optimization problems including linear programming (LP), semi-definite programming (SDP), and a broad class of stochastic gradient descent (SGD) instances. Numerical optimization is a fundamental algorithmic building block: applications include scheduling and resource allocation tasks, approximations...

Deep learning (DL)-based profiled attack has been proved to be a powerful tool in side-channel analysis. A variety of multi-layer perception (MLP) networks and convolutional neural networks (CNN) are thereby applied to cryptographic algorithm implementations for exploiting correct keys with a smaller number of traces and a shorter time. However, most attacks merely focus on small datasets, in which their points of interest are well-trimmed for attacks. Countermeasures applied in embedded...

One of the main promoted advantages of deep learning in profiling side-channel analysis is the possibility of skipping the feature engineering process. Despite that, most recent publications consider feature selection as the attacked interval from the side-channel measurements is pre-selected. This is similar to the worst-case security assumptions in security evaluations when the random secret shares (e.g., mask shares) are known during the profiling phase: an evaluator can identify points...

The majority of recently demonstrated Deep-Learning Side-Channel Analysis (DLSCA) use neural networks trained on a segment of traces containing operations only related to the target subkey. However, when the size of the training set is limited, as in this paper with only 5K power traces, the deep learning (DL) model cannot effectively learn the internal features of the data due to insufficient training data. In this paper, we propose a cross-subkey training approach that acts as a trace...

Recently, the application of multi-party secure computing schemes based on homomorphic encryption in the field of machine learning attracts attentions across the research fields. Previous studies have demonstrated that secure protocols adopting packed additive homomorphic encryption (PAHE) schemes based on the ring learning with errors (RLWE) problem exhibit significant practical merits, and are particularly promising in enabling efficient secure inference in machine-learning-as-a-service...

Computation on ciphertexts of all known fully homomorphic encryption (FHE) schemes induces some noise, which, if too large, will destroy the plaintext. Therefore, the bootstrapping technique that re-encrypts a ciphertext and reduces the noise level remains the only known way of building FHE schemes for arbitrary unbounded computations. The bootstrapping step is also the major efficiency bottleneck in current FHE schemes. A promising direction towards improving concrete efficiency is to...

Deep Learning Side-Channel Attacks (DLSCAs) have become a realistic threat to implementations of cryptographic algorithms, such as Advanced Encryption Standard (AES). By utilizing deep-learning models to analyze side-channel measurements, the attacker is able to derive the secret key of the cryptographic alrgorithm. However, when traces have multiple leakage intervals for a specific attack point, the majority of existing works train neural networks on these traces directly, without a...

Machine learning as a service (MLaaS) has risen to become a prominent technology due to the large development time, amount of data, hardware costs, and level of expertise required to develop a machine learning model. However, privacy concerns prevent the adoption of MLaaS for applications with sensitive data. A promising privacy preserving solution is to use fully homomorphic encryption (FHE) to perform the ML computations. Recent advancements have lowered computational costs by several...

Training deep neural networks requires large scale data, which often forces users to work in a distributed or outsourced setting, accompanied with privacy concerns. Split learning framework aims to address this concern by splitting up the model among the client and the server. The idea is that since the server does not have access to client's part of the model, the scheme supposedly provides privacy. We show that this is not true via two novel attacks. (1) We show that an honest-but-curious...

At CRYPTO'19, Gohr built a bridge between deep learning and cryptanalysis. Based on deep neural networks, he trained neural distinguishers of Speck32/64 using a plaintext difference and single ciphertext pair. Compared with purely differential distinguishers, neural distinguishers successfully use features of the ciphertext pairs. Besides, with the help of neural distinguishers, he attacked 11-round Speck32/64 using Bayesian optimization. At EUROCRYPTO'21, Benamira proposed a detailed...

Profiling attacks, especially those based on machine learning, proved to be very successful techniques in recent years when considering the side-channel analysis of symmetric-key crypto implementations. At the same time, the results for implementations of asymmetric-key cryptosystems are very sparse. This paper considers several machine learning techniques to mount side-channel attacks on two implementations of scalar multiplication on the elliptic curve Curve25519. The first implementation...

In this paper, we show that a software implementation of CCA secure Saber KEM protected by first-order masking and shuffling can be broken by deep learning-based power analysis. Using an ensemble of deep neural networks created at the profiling stage, we can recover the session key and the long-term secret key from $257 \times N$ and $24 \times 257 \times N$ traces, respectively, where $N$ is the number of repetitions of the same measurement. The value of $N$ depends on the implementation,...

Deep learning (DL) has revolutionized Side Channel Analysis (SCA) in recent years. One of the major advantages of DL in the context of SCA is that it can automatically handle masking and desynchronization countermeasures, even while they are applied simultaneously for a cryptographic implementation. However, the success of the attack strongly depends on the DL model used for the attack. Traditionally, Convolutional Neural Networks (CNNs) have been utilized in this regard. This work proposes...

Mixing arithmetic and boolean circuits to perform privacy-preserving machine learning has become increasingly popular. Towards this, we propose a framework for the case of four parties with at most one active corruption called Tetrad. Tetrad works over rings and supports two levels of security, fairness and robustness. The fair multiplication protocol costs 5 ring elements, improving over the state-of-the-art Trident (Chaudhari et al. NDSS'20). A key feature of Tetrad is that robustness...

Machine Learning (ML) algorithms, especially deep neural networks (DNN), have proven themselves to be extremely useful tools for data analysis, and are increasingly being deployed in systems operating on sensitive data, such as recommendation systems, banking fraud detection, and healthcare systems. This underscores the need for privacy-preserving ML (PPML) systems, and has inspired a line of research into how such systems can be constructed efficiently. We contribute to this line of...

Recent progress in interactive zero-knowledge (ZK) proofs has improved the efficiency of proving large-scale computations significantly. Nevertheless, real-life applications (e.g., in the context of private inference using deep neural networks) often involve highly complex computations, and existing ZK protocols lack the expressiveness and scalability to prove results about such computations efficiently. In this paper, we design, develop, and evaluate a ZK system (Mystique) that allows for...