%0 Conference Proceedings
%T Budget-Constrained Result Integrity Verification of Outsourced Data Mining Computations
%+ Stevens Institute of Technology [Hoboken]
%+ Montclair State University [USA] (MSU)
%A Zhang, Bo
%A Dong, Boxiang
%A Wang, Wendy
%Z Part 3: Cloud Security
%< avec comité de lecture
%( Lecture Notes in Computer Science
%B 31th IFIP Annual Conference on Data and Applications Security and Privacy (DBSEC)
%C Philadelphia, PA, United States
%Y Giovanni Livraga
%Y Sencun Zhu
%I Springer International Publishing
%3 Data and Applications Security and Privacy XXXI
%V LNCS-10359
%P 311-324
%8 2017-07-19
%D 2017
%R 10.1007/978-3-319-61176-1_17
%K Data-Mining-as-a-Service (DMaS)
%K Cloud computing
%K Result integrity
%K Budgeted maximization
%Z Computer Science [cs]Conference papers
%X When outsourcing data mining needs to an untrusted service provider in the Data-Mining-as-a-Service (DMaS) paradigm, it is important to verify whether the service provider (server) returns correct mining results (in the format of data mining objects). We consider the setting in which each data mining object is associated with a weight for its importance. Given a client who is equipped with limited verification budget, the server selects a subset of mining results whose total verification cost does not exceed the given budget, while the total weight of the selected results is maximized. This maps to the well-known budgeted maximum coverage (BMC) problem, which is NP-hard. Therefore, the server may execute a heuristic algorithm to select a subset of mining results for verification. The server has financial incentives to cheat on the heuristic output, so that the client has to pay more for verification of the mining results that are less important. Our aim is to verify that the mining results selected by the server indeed satisfy the budgeted maximization requirement. It is challenging to verify the result integrity of the heuristic algorithms as the results are non-deterministic. We design a probabilistic verification method by including negative candidates (NCs) that are guaranteed to be excluded from the budgeted maximization result of the ratio-based BMC solutions. We perform extensive experiments on real-world datasets, and show that the NC-based verification approach can achieve high guarantee with small overhead.
%G English
%Z TC 11
%Z WG 11.3
%2 https://inria.hal.science/hal-01684361/document
%2 https://inria.hal.science/hal-01684361/file/453481_1_En_17_Chapter.pdf
%L hal-01684361
%U https://inria.hal.science/hal-01684361
%~ IFIP-LNCS
%~ IFIP
%~ IFIP-TC
%~ IFIP-WG
%~ IFIP-TC11
%~ IFIP-WG11-3
%~ IFIP-DBSEC
%~ IFIP-LNCS-10359