International Association
for Cryptologic Research

We consider the problem of verifiable and private delegation of computation [Gennaro et al. CRYPTO'10] in which a client stores private data on an untrusted server and asks the server to compute functions over this data. In this scenario we aim to achieve three main properties: the server should not learn information on inputs and outputs of the computation (privacy), the server cannot return wrong results without being caught (integrity), and the client can verify the correctness of the outputs faster than running the computation (efficiency). A known paradigm to solve this problem is to use a (non-private) verifiable computation (VC) to prove correctness of a homomorphic encryption (HE) evaluation on the ciphertexts. Despite the research advances in obtaining efficient VC and HE, using these two primitives together in this paradigm is concretely expensive. Recent work [Fiore et al. CCS'14, PKC'20] addressed this problem by designing specialized VC solutions that however require the HE scheme to work with very specific parameters; notably HE ciphertexts must be over $\mathbb{Z}_q$ for a large prime $q$. In this work we propose a new solution that allows a flexible choice of HE parameters, while staying modular (based on the paradigm combining VC and HE) and efficient (the VC and the HE schemes are both executed at their best efficiency). At the core of our new protocol are new homomorphic hash functions for Galois rings. As an additional contribution we extend our results to support non-deterministic computations on encrypted data and an additional privacy property by which verifiers do not learn information on the inputs of the computation.

We propose a multi-party computation (MPC) protocol over $\mathbb{Z}_{2^k}$ secure against actively corrupted majority from somewhat homomorphic encryption. The main technical contributions are: (i) a new efficient packing method for $\mathbb{Z}_{2^k}$-messages in lattice-based somewhat homomorphic encryption schemes, (ii) a simpler reshare protocol for level-dependent packings, (iii) a more efficient zero-knowledge proof of plaintext knowledge on cyclotomic rings $\Z[X]/\Phi_M(X)$ with $M$ being a prime. Integrating them, our protocol shows from 2.2x upto 4.8x improvements in amortized communication costs compared to the previous best results. Our techniques not only improve the efficiency of MPC over $\mathbb{Z}_{2^k}$ considerably, but also provide a toolkit that can be leveraged when designing other cryptographic primitives over $\mathbb{Z}_{2^k}$.

Comparison of two numbers is one of the most frequently used operations, but it has been a challenging task to efficiently compute the comparison function in homomorphic encryption~(HE) which basically support addition and multiplication. Recently, Cheon et al.~(Asiacrypt~2019) introduced a new approximate representation of the comparison function with a rational function, and showed that this rational function can be evaluated by an iterative algorithm. Due to this iterative feature, their method achieves a logarithmic computational complexity compared to previous polynomial approximation methods; however, the computational complexity is still not optimal, and the algorithm is quite slow for large-bit inputs in HE implementation. In this work, we propose new comparison methods with \emph{optimal} asymptotic complexity based on \emph{composite polynomial} approximation. The main idea is to systematically design a constant-degree polynomial $f$ by identifying the \emph{core properties} to make a composite polynomial $f\circ f \circ \cdots \circ f$ get close to the sign function (equivalent to the comparison function) as the number of compositions increases. We additionally introduce an acceleration method applying a mixed polynomial composition $f\circ \cdots \circ f\circ g \circ \cdots \circ g$ for some other polynomial $g$ with different properties instead of $f\circ f \circ \cdots \circ f$. Utilizing the devised polynomials $f$ and $g$, our new comparison algorithms only require $\Theta(\log(1/\epsilon)) + \Theta(\log\alpha)$ computational complexity to obtain an approximate comparison result of $a,b\in[0,1]$ satisfying $|a-b|\ge \epsilon$ within $2^{-\alpha}$ error. The asymptotic optimality results in substantial performance enhancement: our comparison algorithm on $16$-bit encrypted integers for $\alpha = 16$ takes $1.22$ milliseconds in amortized running time based on an approximate HE scheme HEAAN, which is $18$ times faster than the previous work.

We propose a new method to compare numbers which are encrypted by Homomorphic Encryption (HE). Previously, comparison and min/max functions were evaluated using Boolean functions where input numbers are encrypted bit-wise. However, the bit-wise encryption methods require relatively expensive computations for basic arithmetic operations such as addition and multiplication.In this paper, we introduce iterative algorithms that approximately compute the min/max and comparison operations of several numbers which are encrypted word-wise. From the concrete error analyses, we show that our min/max and comparison algorithms have $$\varTheta (\alpha )$$ and $$\varTheta (\alpha \log \alpha )$$ computational complexity to obtain approximate values within an error rate $$2^{-\alpha }$$, while the previous minimax polynomial approximation method requires the exponential complexity $$\varTheta (2^{\alpha /2})$$ and $$\varTheta (\sqrt{\alpha }\cdot 2^{\alpha /2})$$, respectively. Our algorithms achieve (quasi-)optimality in terms of asymptotic computational complexity among polynomial approximations for min/max and comparison operations. The comparison algorithm is extended to several applications such as computing the top-k elements and counting numbers over the threshold in encrypted state.Our method enables word-wise HEs to enjoy comparable performance in practice with bit-wise HEs for comparison operations while showing much better performance on polynomial operations. Computing an approximate maximum value of any two $$\ell $$-bit integers encrypted by HEAAN, up to error $$2^{\ell -10}$$, takes only 1.14 ms in amortized running time, which is comparable to the result based on bit-wise HEs.