Machine learning relies on the availability of a vast amount of data for training. However, in reality, most data are scattered across different organizations and cannot be easily integrated under many legal and practical constraints. In this paper, we introduce a new technique and framework, known as federated transfer learning (FTL), to improve statistical models under a data federation. The federation allows knowledge to be shared without compromising user privacy, and enables complimentary knowledge to be transferred in the network. As a result, a target-domain party can build more flexible and powerful models by leveraging rich labels from a source-domain party. A secure transfer cross validation approach is also proposed to guard the FTL performance under the federation. The framework requires minimal modifications to the existing model structure and provides the same level of accuracy as the nonprivacy-preserving approach. This framework is very flexible and can be effectively adapted to various secure multi-party machine learning tasks.

No abstract

We construct digital (t, s)-sequences in a prime-power base q for which the quality parameter t has the least possible order of magnitude. The construction uses algebraic function fields over the finite field of order q which contain many places of degree 1 relative to the genus.

In this paper, we first construct several classes of classical Hermitian self-orthogonal MDS codes. Through these classical codes, we are able to obtain various quantum MDS codes. It turns out that many of our quantum codes are new in the sense that the parameters of our quantum codes cannot be obtained from all previous constructions.

Most multi-party computation protocols allow secure computation of arithmetic circuits over a finite field, such as the integers modulo a prime. In the more natural setting of integer computations modulo 2 k , which are useful for simplifying implementations and applications, no solutions with active security are known unless the majority of the participants are honest.We present a new scheme for information-theoretic MACs that are homomorphic modulo 2 k , and are as efficient as the well-known standard solutions that are homomorphic over fields. We apply this to construct an MPC protocol for dishonest majority in the preprocessing model that has efficiency comparable to the well-known SPDZ protocol (Damgård et al., CRYPTO 2012), with operations modulo 2 k instead of over a field. We also construct a matching preprocessing protocol based on oblivious transfer, which is in the style of the MASCOT protocol (Keller et al., CCS 2016) and almost as efficient.