Abstract
We propose Soteria, a system for distributed privacy-preserving Machine Learning (ML) that leverages Trusted Execution Environments (e.g. Intel SGX) to run code in isolated containers (enclaves). Unlike previous work, where all ML-related computation is performed at trusted enclaves, we introduce a hybrid scheme, combining computation done inside and outside these enclaves. The conducted experimental evaluation validates that our approach reduces the runtime of ML algorithms by up to 41%, when compared to previous related work. Our protocol is accompanied by a security proof, as well as a discussion regarding resilience against a wide spectrum of ML attacks.

Abstract

Abstract

Abstract

Abstract
Conflict-free Replicated Data Types (CRDTs) are abstract data types that support developers when designing and reasoning about distributed systems with eventual consistency guarantees. In their core they solve the problem of how to deal with concurrent operations, in a way that is transparent for developers. However in the real world, distributed systems also suffer from other relevant problems, including security and privacy issues and especially when participants can be untrusted. In this paper we present new privacy-preserving CRDT protocols that can be used to help secure distributed cloud-backed applications, including NoSQL geo-replicated databases. Our proposals are based on standard CRDTs, such as sets and counters, augmented with cryptographic mechanisms that allow their operations to be performed on encrypted data. We accompany our proposals with formal security proofs and implement and integrate them in An-tidoteDB, a geo-replicated NoSQL database that leverages CRDTs for its operations. Experimental evaluations based on the Danish Shared Medication Record dataset (FMK) exhibit the tradeoffs that our different proposals make and show that they are ready to be used in practical applications.

Abstract
In this paper we show how Isolated Execution Environments (IEE) offered by novel commodity hardware such as Intel’s SGX provide a new path to constructing general secure multiparty computation (MPC) protocols. Our protocol is intuitive and elegant: it uses code within an IEE to play the role of a trusted third party (TTP), and the attestation guarantees of SGX to bootstrap secure communications between participants and the TTP. The load of communications and computations on participants only depends on the size of each party’s inputs and outputs and is thus small and independent from the intricacies of the functionality to be computed. The remaining computational load– essentially that of computing the functionality – is moved to an untrusted party running an IEE-enabled machine, an attractive feature for Cloud-based scenarios. Our rigorous modular security analysis relies on the novel notion of labeled attested computation which we put forth in this paper. This notion is a convenient abstraction of the kind of attestation guarantees one can obtain from trusted hardware in multi-user scenarios. Finally, we present an extensive experimental evaluation of our solution on SGX-enabled hardware. Our implementation is open-source and it is functionality agnostic: it can be used to securely outsource to the Cloud arbitrary off-the-shelf collaborative software, such as the one employed on financial data applications, enabling secure collaborative execution over private inputs provided by multiple parties. © 2017, International Financial Cryptography Association.