International Association
for Cryptologic Research

The rise of machine learning as a service multiplies scenarios where one faces a privacy dilemma: either sensitive user data must be revealed to the entity that evaluates the cognitive model (e.g., in the Cloud), or the model itself must be revealed to the user so that the evaluation can take place locally. Fully Homomorphic Encryption (FHE) offers an elegant way to reconcile these conflicting interests in the Cloud-based scenario and also preserve non-interactivity. However, due to the inefficiency of existing FHE schemes, most applications prefer to use Somewhat Homomorphic Encryption (SHE), where the complexity of the computation to be performed has to be known in advance, and the efficiency of the scheme depends on this global complexity.In this paper, we present a new framework for homomorphic evaluation of neural networks, that we call FHE–DiNN, whose complexity is strictly linear in the depth of the network and whose parameters can be set beforehand. To obtain this scale-invariance property, we rely heavily on the bootstrapping procedure. We refine the recent FHE construction by Chillotti et al. (ASIACRYPT 2016) in order to increase the message space and apply the sign function (that we use to activate the neurons in the network) during the bootstrapping. We derive some empirical results, using TFHE library as a starting point, and classify encrypted images from the MNIST dataset with more than 96% accuracy in less than 1.7 s.Finally, as a side contribution, we analyze and introduce some variations to the bootstrapping technique of Chillotti et al. that offer an improvement in efficiency at the cost of increasing the storage requirements.

We identify and fill some gaps with regard to consistency (the extent to which false positives are produced) for public-key encryption with keyword search (PEKS). We define computational and statistical relaxations of the existing notion of perfect consistency, show that the scheme of Boneh et al. in Eurocrypt 2004 is computationally consistent, and provide a new scheme that is statistically consistent. We also provide a transform of an anonymous IBE scheme to a secure PEKS scheme that, unlike the previous one, guarantees consistency. Finally, we suggest three extensions of the basic notions considered here, namely anonymous HIBE, public-key encryption with temporary keyword search, and identity-based encryption with keyword search.

This paper presents the theoretical blueprint of a new secure token called the Externalized Microprocessor (XmP). Unlike a smart-card, the XmP contains no ROM at all. While exporting all the device's executable code to potentially untrustworthy terminals poses formidable security problems, the advantages of ROM-less secure tokens are numerous: chip masking time disappears, bug patching becomes a mere terminal update and hence does not imply any roll-out of cards in the field. Most importantly, code size ceases to be a limiting factor. This is particularly significant given the steady increase in on-board software complexity. After describing the machine's instruction-set we will introduce two XmP variants. The first design is a public-key oriented architecture which relies on a new RSA screening scheme and features a relatively low communication overhead at the cost of computational complexity, whereas the second variant is secret-key oriented and relies on simple MACs and hash functions but requires more communication. For each of these two designs, we propose two protocols that execute and dynamically authenticate arbitrary programs. We also provide a strong security model for these protocols and prove their security under appropriate complexity assumptions.

This paper presents a new attack on keyboards. \smallskip The attack consists in depositing on each keyboard key a small ionic salt quantity ({\sl e.g.} some NaCl on key 0, some KCl on key 1, LiCl on key 2, SrCl$_2$ on key 3, BaCl$_2$ on key 4, CaCl$_2$ on key 5...). As the user enters his PIN, salts get mixed and leave the keyboard in a state that leaks secret information. Nicely enough, evaluating the entropy loss due to the chemical trace turns out to be a very interesting combinatorial exercise. \smallskip Under the assumption that mass spectroscopic analysis can reveal with accuracy the mixture of chemical compounds generated by the user, we show that, for moderate-size decimal PINs, the attack would generally disclose the PIN. \smallskip The attack may apply to door PIN codes, phone numbers dialed from a hotel rooms, computer keyboards or even ATMs. \ss While we did not implement the chemical part of the attack, a number of mass spectrometry specialists confirmed to the authors its feasibility.

This paper considers arbitrary-length chosen-ciphertext secure asymmetric encryption, thus addressing what is actually needed for a practical usage of strong public-key cryptography in the real world. We put forward two generic constructions, gem-1 and gem-2, which apply to explicit fixed-length weakly secure primitives and provide a strongly secure (IND-CCA2) public-key encryption scheme for messages of unfixed length (typically computer files). Our techniques optimally combine a single call to any one-way trapdoor function with repeated encryptions through some weak block-cipher (a simple xor is fine) and hash functions of fixed-length input so that a minimal number of calls to these functions is needed. Our encryption/decryption throughputs are comparable to the ones of standard methods (asymmetric encryption of a session key + symmetric encryption with multiple modes). In our case, however, we formally prove that our designs are secure in the strongest sense and provide complete security reductions holding in the random oracle model.

A common practice to encrypt with RSA is to first apply a padding scheme to the message and then to exponentiate the result with the public exponent; an example of this is OAEP. Similarly, the usual way of signing with RSA is to apply some padding scheme and then to exponentiate the result with the private exponent, as for example in PSS. Usually, the RSA modulus used for encrypting is different from the one used for signing. The goal of this paper is to simplify this common setting. First, we show that PSS can also be used for encryption, and gives an encryption scheme semantically secure against adaptive chosen-ciphertext attacks, in the random oracle model. As a result, PSS can be used indifferently for encryption or signature. Moreover, we show that PSS allows to safely use the same RSA key-pairs for both encryption and signature, in a concurrent manner. More generally, we show that using PSS the same set of keys can be used for both encryption and signature for any trapdoor partial-domain one-way permutation. The practical consequences of our result are important: PKIs and public-key implementations can be significantly simplified.