Workshop on Evolutionary Algorithms for Problems with Uncertainty

July 8th-12th 2020

Call for Papers

In many real-world optimisation problems, uncertainty is present in various forms. One prominent example is the sensitivity of the optimal solution to noise or perturbations in the environment. In such cases, handling uncertainty effectively can be critical for finding good robust solutions, in particular, when the uncertainty results in severe loss of quality. In recent years, uncertainty in its various forms has attracted a lot of attention from the evolutionary computation community.

Optimisation problems can be categorised as one of four types, depending on the source of uncertainty:

  • robust problems, where the uncertainty arises in design or environmental variables,
  • noisy problems, where the uncertainty arises in objective space,
  • approximated problems, where approximated objective function(s) are subject to error, and
  • dynamic problems, where the objective function(s) changes over time.
  • Robust optimisation includes situations where the chosen design cannot be realised in a real-world setting without some error. Additionally, the solution may need to perform well under a set of different scenarios and/or under some assumptions of parameter drifts. Typically, explicit methods for handling this type of uncertainty rely on resampling the assumed scenario set in order to approximate the underlying robust fitness landscape. Noisy optimisation refers to problems in which the estimate of the quality of an individual is subject to some randomness, e.g. if the objective value is calculated from the output of a stochastic simulation or solver. In this case, the estimate of the expected objective value is usually based on several resamples of a given solution. However, methods that rely on resampling of solutions are often inadequate in situations where the evaluations are expensive.

    These problems have been a concern for the community for a number of years, and there is a growing need for new methods to handle the various types of uncertainty in a wide variety of problem domains. In addition, the field stands to benefit greatly from new methods for assessing the performance of algorithms for optimisation in uncertain environments and development of suitable benchmark problems. This workshop is designed to bring together practitioners from different subfields in the evolutionary computing community to share their ideas and methods.

    Particular topics of interest include, but are not limited to:

    • Efficient methods for optimisation under uncertainty
    • Studies of the inherent capabilities of EAs to handle different types of uncertainty
    • New ranking and selections operators for optimising under uncertainty
    • Meta-modelling for handling uncertainty
    • Methods for fitness approximation under uncertainty
    • Quantifying the robustness of solutions
    • Real-World applications that suffer from various types of uncertainty
    • New benchmark problems for various types of uncertainty
    • Design of experiments for estimating robust designs
    • Coping with multiple sources and forms of uncertainty
    • Multi-objective optimisation in uncertain contexts
    • Casting a problem with uncertainty as a multi-objective problem

    Submission

    Authors should refer to the GECCO Submission Instructions. The abstract must not exceed 200 words and the maximum length of papers is 8 pages (excluding references). We also welcome position papers of up to 2 pages showcasing exciting exploratory and preliminary results. Accepted papers will be published in the GECCO Companion Proceedings and will be presented orally at the workshop. Please note that submissions for GECCO workshops will be handled differently to previous years. All submissions will now be handled through the GECCO Submission Site.

    Acknowledgement

    This workshop is supported by the Engineering and Physical Sciences Research Council, UK [grant numbers: EP/N017846/1, EP/N014391/1].