We propose an computational framework for real-time risk assessment and prioritizing for random outcomes without prior information on probability distributions. The basic model is built based on satisficing measure (SM) which yields a single index for risk comparison. Since SM is a dual representation for a family of risk measures, we consider problems constrained by general convex risk measures and specifically by Conditional value-at-risk. Starting from offline optimization, we apply sample average approximation technique and argue the convergence rate and validation of optimal solutions. In online stochastic optimization case, we develop primal-dual stochastic approximation algorithms respectively for general risk constrained problems, and derive their regret bounds. For both offline and online cases, we illustrate the relationship between risk ranking accuracy with sample size (or iterations). Sandoy and Aven (2006) can solve problems generated by introducing or not introducing an underlying probability model. The pairwise comparison theory (also called Bradley-Terry-Luce (BTL) model) widely used in study the preference in decision making was introduced in Bradley and Terry (1952); Luce (2005). A Bayesian approximation method with BTL model in Weng and Lin (2011), is proposed for online ranking in team performance. However, these methodologies do not consider assessing the "risk" of the systems and have not provided a formal definition of risk that can be uniformly applied to wide range of systems.
Bayesian approaches used inIn financial mathematics, "risk measure" is defined as a mapping function from a probability space to real number. Some fundamental research has been conducted motivating the definition of risk measure. In Ruszczynski and Shapiro (2006), the definitions and conditions for convex and coherent risk measures are developed, and conjugate duality reformulation of risk functions is proposed. Rockafellar and Uryasev (2000) give the detailed arguments on the most widely investigated coherent and law invariant risk measure-Conditional value-at-risk (CVaR) with corresponding reformulation and optimization problem illustration. CVaR is widely involved in optimization under uncertainty, and helps to improve the reliability of solutions against extremely high loss. Krokhmal et al. (2002) study the portfolio optimization problem with CVaR objective and constraints, and corresponding reformulation and discretization are demonstrated. In Dai et al. (2016), robust version CVaR is applied in portfolio selection problem, where sampled scenario returns are generated by a factor model with some asymmetric and uncertainty set. In Noyan and Rudolf (2013), multivariate CVaR constraint problem is studied based on polyhedral scalarization and second-order stochastic dominance, and a cut generation algorithm is proposed, where each cut is obtained by solving a mixed integer problem. Xu and Yu (2014) reformulate a stochastic nonlinear complementary problem with CVaR constraints, and propose a penalized smoothing sam...