Threshold implementation is a method based on secret sharing to secure cryptographic ciphers (and in particular S-boxes) against differential power analysis side-channel attacks which was proposed by Nikova, Rechberger, and Rijmen in 2006. Until now, threshold implementations were only constructed for specific types of functions and some small S-boxes, but no generic construction was ever presented. In this paper, we present the first universal threshold implementation with$t+2$shares that is applicable to any bijective S-box, where$t$is its algebraic degree (or is larger than the algebraic degree). While being universal, our construction is also optimal with respect to the number of shares, since the theoretically smallest possible number,$t+1$, is not attainable for some bijective S-boxes. Our results enable low latency secure hardware implementations without the need for additional randomness. In particular, we apply this result to find two uniform sharings of the AES S-box. The first sharing is obtained by using the threshold implementation of the inversion in$\mathbb {F}_{2^{8}}$and the second by using two threshold implementations of two cubic power permutations that decompose the inversion. Area and performance figures for hardware implementations are provided.

Abstract Over the years, a plethora of cost-sensitive methods have been proposed for learning on data when different types of misclassification errors incur different costs. Our contribution is a unifying framework that provides a comprehensive and insightful overview on cost-sensitive ensemble methods, pinpointing their differences and similarities via a fine-grained categorization. Our framework contains natural extensions and generalisations of ideas across methods, be it AdaBoost, Bagging or Random Forest, and as a result not only yields all methods known to date but also some not previously considered.

This paper describes a decision support system designed for a Belgian Human Resource (HR) and Well-Being Service Provider. Their goal is to improve health and well-being in the workplace, and to this end, the task is to identify groups of employees at risk of sickness absence who can then be targeted with interventions aiming to reduce or prevent absences. To facilitate deployment, we apply a range of existing machine-learning methods to obtain predictions at monthly intervals using real HR and payroll data that contains no health-related predictors. We model employee absence as a binary classification problem with loss asymmetry and conceptualise a misclassification cost matrix of employee sickness absence. Model performance is evaluated using cost-based metrics, which have intuitive interpretation. We also demonstrate how this problem can be approached when costs are unknown. The proposed flexible evaluation procedure is not restricted to a specific model or domain and can be applied to address other HR analytics questions when deployed. Our approach of considering a wider range of methods and cost-based performance evaluation is novel in the domain of absenteeism prediction.