Karn Seth

Indistinguishability Obfuscation from Semantically-Secure Multilinear Encodings

We define a notion of semantic security of multilinear (a.k.a. graded) encoding schemes, which stipulates security of a class of algebraic “decisional” assumptions: roughly speaking, we require that for every nuPPT distribution D over two constant-length sequences m0,m1 and auxiliary elements z such that all arithmetic circuits (respecting the multilinear restrictions and ending with a zero-test) are constant with overwhelming probability over (m b , z), b ∈ {0,1}, we have that encodings of m0, z are computationally indistinguishable from encodings of m1, z. Assuming the existence of semantically secure multilinear encodings and the LWE assumption, we demonstrate the existence of indistinguishability obfuscators for all polynomial-size circuits.

Practical Secure Aggregation for Privacy-Preserving Machine Learning

We design a novel, communication-efficient, failure-robust protocol for secure aggregation of high-dimensional data. Our protocol allows a server to compute the sum of large, user-held data vectors from mobile devices in a secure manner (i.e. without learning each users individual contribution), and can be used, for example, in a federated learning setting, to aggregate user-provided model updates for a deep neural network. We prove the security of our protocol in the honest-but-curious and active adversary settings, and show that security is maintained even if an arbitrarily chosen subset of users drop out at any time. We evaluate the efficiency of our protocol and show, by complexity analysis and a concrete implementation, that its runtime and communication overhead remain low even on large data sets and client pools. For 16-bit input values, our protocol offers

On the impossibility of cryptography with tamperable randomness

1.73 x communication expansion for 210 users and 220-dimensional vectors, and 1.98 x expansion for 214 users and 224-dimensional vectors over sending data in the clear.

On the Impossibility of Black-Box Transformations in Mechanism Design

We initiate a study of the security of cryptographic primitives in the presence of efficient tampering attacks to the randomness of honest parties. More precisely, we consider p-tampering attackers that may efficiently tamper with each bit of the honest parties’ random tape with probability p, but have to do so in an “online” fashion. Our main result is a strong negative result: We show that any secure encryption scheme, bit commitment scheme, or zero-knowledge protocol can be “broken” with probability p by a p-tampering attacker.The core of this result is a new Fourier analytic technique for biasing the output of bounded-value functions, which may be of independent interest.

Output-Compressing Randomized Encodings and Applications

A fundamental question in algorithmic mechanism design is whether any approximation algorithm for a single-parameter social-welfare maximization problem can be turned into a dominant-strategy truthful mechanism for the same problem (while preserving the approximation ratio up to a constant factor). A particularly desirable type of transformations—called black-box transformations—achieve the above goal by only accessing the approximation algorithm as a black box.