Matrix Factorization Techniques for Recommender Systems

This paper shows a generic and simple conversion from weak asymmetric and symmetric encryption schemes into an asymmetric encryption scheme which is secure in a very strong sense- indistinguishability against adaptive chosen-ciphertext attacks in the random oracle model. In particular, this conversion can be applied efficiently to an asymmetric encryption scheme that provides a large enough coin space and, for every message, many enough variants of the encryption, like the ElGamal encryption scheme.

Secure integration of asymmetric and symmetric encryption schemes

This paper studies model-inversion attacks, in which the access to a model is abused to infer information about the training data. Since its first introduction by~\cite{fredrikson2014privacy}, such attacks have raised serious concerns given that training data usually contain privacy sensitive information. Thus far, successful model-inversion attacks have only been demonstrated on simple models, such as linear regression and logistic regression. Previous attempts to invert neural networks, even the ones with simple architectures, have failed to produce convincing results. Here we present a novel attack method, termed the \emph{generative model-inversion attack}, which can invert deep neural networks with high success rates. Rather than reconstructing private training data from scratch, we leverage partial public information, which can be very generic, to learn a distributional prior via generative adversarial networks (GANs) and use it to guide the inversion process. Moreover, we theoretically prove that a model's predictive power and its vulnerability to inversion attacks are indeed two sides of the same coin---highly predictive models are able to establish a strong correlation between features and labels, which coincides exactly with what an adversary exploits to mount the attacks. Our extensive experiments demonstrate that the proposed attack improves identification accuracy over the existing work by about $75\%$ for reconstructing face images from a state-of-the-art face recognition classifier. We also show that differential privacy, in its canonical form, is of little avail to defend against our attacks.

/pdf/the-secret-revealer-generative-model-inversion-attacks-4bizqps1ct.pdf

The Secret Revealer: Generative Model-Inversion Attacks Against Deep Neural Networks

We construct two efficient Identity-Based Encryption (IBE) systems that admit selective-identity security reductions without random oracles in groups equipped with a bilinear map. Selective-identity secure IBE is a slightly weaker security model than the standard security model for IBE. In this model the adversary must commit ahead of time to the identity that it intends to attack, whereas in an adaptive-identity attack the adversary is allowed to choose this identity adaptively. Our first system—BB1—is based on the well studied decisional bilinear Diffie–Hellman assumption, and extends naturally to systems with hierarchical identities, or HIBE. Our second system—BB2—is based on a stronger assumption which we call the Bilinear Diffie–Hellman Inversion assumption and provides another approach to building IBE systems.

Our first system, BB1, is very versatile and well suited for practical applications: the basic hierarchical construction can be efficiently secured against chosen-ciphertext attacks, and further extended to support efficient non-interactive threshold decryption, among others, all without using random oracles. Both systems, BB1 and BB2, can be modified generically to provide “full” IBE security (i.e., against adaptive-identity attacks), either using random oracles, or in the standard model at the expense of a non-polynomial but easy-to-compensate security reduction.

Efficient selective identity-based encryption without random oracles

Efficient implementations of lattice-based cryptographic schemes have been limited to only the most basic primitives like encryption and digital signatures. The main reason for this limitation is that at the core of many advanced lattice primitives is a trapdoor sampling algorithm (Gentry, Peikert, Vaikuntanathan, STOC 2008) that produced outputs that were too long for practical applications. In this work, we show that using a particular distribution over NTRU lattices can make GPV-based schemes suitable for practice. More concretely, we present the first lattice-based IBE scheme with practical parameters – key and ciphertext sizes are between two and four kilobytes, and all encryption and decryption operations take approximately one millisecond on a moderately-powered laptop. As a by-product, we also obtain digital signature schemes which are shorter than the previously most-compact ones of Ducas, Durmus, Lepoint, and Lyubashevsky from Crypto 2013.

/pdf/e-fficient-identity-based-encryption-over-ntru-lattices-1anjd774p8.pdf

E fficient Identity-Based Encryption over NTRU Lattices.

In this paper, we present new adaptively secure identity-based encryption IBE schemes. One of the distinguishing properties of the schemes is that it achieves shorter public parameters than previous schemes. Both of our schemes follow the general framework presented in the recent IBE scheme of Yamada Eurocrypt 2016, employed with novel techniques tailored to meet the underlying algebraic structure to overcome the difficulties arising in our specific setting. Specifically, we obtain the following:

- Our first scheme is proven secure under the ring learning with errors RLWE assumption and achieves the best asymptotic space efficiency among existing schemes from the same assumption. The main technical contribution is in our new security proof that exploits the ring structure in a crucial way. Our technique allows us to greatly weaken the underlying hardness assumption e.g., we assume the hardness of RLWE with a fixed polynomial approximation factor whereas Yamada's scheme requires a super-polynomial approximation factor while improving the overall efficiency.

- Our second IBE scheme is constructed on bilinear maps and is secure under the 3-computational bilinear Diffie-Hellman exponent assumption. This is the first IBE scheme based on the hardness of a computational/search problem, rather than a decisional problem such as DDH and DLIN on bilinear maps with sub-linear public parameter size.

Partitioning via Non-linear Polynomial Functions: More Compact IBEs from Ideal Lattices and Bilinear Maps

While online services based on machine learning (ML) have been attracting considerable attention in both academic and business, privacy issues are becoming a threat that cannot be ignored. Recently, Fredrikson et al. [USENIX 2014] proposed a new paradigm of model inversion attacks, which allows an adversary to expose the sensitive information of users by using an ML system for an unintended purpose. In particular, the attack reveals the sensitive attribute values of the target user by using their non-sensitive attributes and the output of the ML model. Here, for the attack to succeed, the adversary needs to possess the non-sensitive attribute values of the target user prior to the attack. However, in reality, even if this information (i.e., non-sensitive attributes) is not necessarily information the user regards as sensitive, it may be difficult for the adversary to actually acquire it. In this paper, we propose a general model inversion (GMI) framework to capture the above scenario where knowledge of the non-sensitive attributes is not necessarily provided. Here, our framework also captures the scenario of Fredrikson et al. Notably, we generalize the paradigm of Fredrikson et al. by additionally modeling the amount of auxiliary information the adversary possesses at the time of the attack. Our proposed GMI framework enables a new type of model inversion attack for prediction systems, which can be carried out without knowledge of the non-sensitive attributes. At a high level, we use the paradigm of data poisoning in a novel way and inject malicious data into the set of training data to modify the ML model into a target ML model, which we can attack without having to have knowledge of the non-sensitive attributes. Our new attack enables the inference of sensitive attributes in the user input from only the output of the ML model, even when the non-sensitive attributes of the user are not available to the adversary. Finally, we provide a concrete algorithm of our model inversion attack on prediction systems based on linear regression models, and give a detailed description of how the data poisoning algorithm is constructed.We evaluate the performance of our new model inversion attack without the knowledge of non-sensitive attributes through experiments with actual data sets.

Model Inversion Attacks for Prediction Systems: Without Knowledge of Non-Sensitive Attributes

We construct efficient ring signatures (RS) from isogeny and lattice assumptions Our ring signatures are based on a logarithmic OR proof for group actions We instantiate this group action by either the CSIDH group action or an MLWE-based group action to obtain our isogeny-based or lattice-based RS scheme, respectively Even though the OR proof has a binary challenge space and therefore requires a number of repetitions which is linear in the security parameter, the sizes of our ring signatures are small and scale better with the ring size N than previously known post-quantum ring signatures We also construct linkable ring signatures (LRS) that are almost as efficient as the non-linkable variants The isogeny-based scheme produces signatures whose size is an order of magnitude smaller than all previously known logarithmic post-quantum ring signatures, but it is relatively slow (eg 55 KB signatures and 79 s signing time for rings with 8 members) In comparison, the lattice-based construction is much faster, but has larger signatures (eg 30 KB signatures and 90 ms signing time for the same ring size) For small ring sizes our lattice-based ring signatures are slightly larger than state-of-the-art schemes, but they are smaller for ring sizes larger than $N \approx 1024$

/pdf/calamari-and-falafl-logarithmic-linkable-ring-signatures-136yfvx1gz.pdf

Calamari and Falafl: Logarithmic (Linkable) Ring Signatures from Isogenies and Lattices

Revocable identity-based encryption (RIBE) is an extension of IBE that supports a key revocation mechanism, which is an indispensable feature for practical cryptographic schemes. Due to this extra feature, RIBE is often required to satisfy a strong security notion unique to the revocation setting called decryption key exposure resistance (DKER). Additionally, hierarchal IBE (HIBE) is another orthogonal extension of IBE that supports key delegation functionalities allowing for scalable deployments of cryptographic schemes. So far, R(H)IBE constructions with DKER are only known from bilinear maps, where all constructions rely heavily on the so-called key re-randomization property to achieve the DKER and/or hierarchal feature. Since lattice-based schemes seem to be inherently ill-fit with the key re-randomization property, no construction of lattice-based R(H)IBE schemes with DKER are known.

/pdf/lattice-based-revocable-hierarchical-ibe-with-decryption-key-3rs77qp4bi.pdf

Lattice-Based Revocable (Hierarchical) IBE with Decryption Key Exposure Resistance.

https://link.springer.com/content/pdf/10.1007%2F978-3-030-17259-6_15.pdf

Shuichi Katsumata

Papers

Partitioning via Non-linear Polynomial Functions: More Compact IBEs from Ideal Lattices and Bilinear Maps

Model Inversion Attacks for Prediction Systems: Without Knowledge of Non-Sensitive Attributes

Calamari and Falafl: Logarithmic (Linkable) Ring Signatures from Isogenies and Lattices

Lattice-Based Revocable (Hierarchical) IBE with Decryption Key Exposure Resistance.

Lattice-based revocable (hierarchical) IBE with decryption key exposure resistance