PPSN 2024 Workshop: 30 Years of EDAs

Call for abstracts now opens. See details.

Description

Estimation of Distribution Algorithms (EDAs) have been an active and impactful research area now for around thirty years. The publication of Population Based Incremental Learning (PBIL) in 1994, crystallised research focused around estimating and sampling models of the distribution of value in a solution space as an operating mechanism for population-based metaheuristic search. Over the last 30 years, EDAs have developed from building univariate (fully-factorised) probabilistic models on simple theoretical problems (OneMax, Trapk, etc) to pervasive application across most problem representation spaces, solving challenging benchmark datasets, demonstrating competitive performance on single- and multi-objective and dynamic problems and used on induction of machine learning models to address supervised learning, feature selection, clustering and reinforcement learning problems as well as on a wide range of real world applications including bioinformatics, biomedicine, computational neuroscience, industry 4.0, evolvable hardware design and quantum computing. Journals like Evolutionary Computation, IEEE Transactions on Evolutionary Computation or International Journal of Approximate Reasoning have organised special issues on EDAs. Conferences have shown an interest in hosting tutorials on EDAs, since the first seminal one at PPSN VI in 2000 to the recent presentation related to theory at this PPSN.

More recently EDAs have become a focus of interest for foundational theoretical research on population-based metaheuristics. The explicit mathematical nature of the EDA update mechanism lends itself to theoretical examination using the established tools of run time analysis as well as EDA-specific analysis methods. This led to several deep insights on the working principles of EDAs, e.g., why they cope well with noise, how they leave local optima faster than traditional evolutionary algorithms, and how the parameters lead to the undesirable effect of genetic drift.

The purpose of this workshop is to mark the thirty year milestone in a mature and diverse topic of continuing interest and to encourage thinking on key open questions and future directions. We invite participants to contribute short talks aimed at stimulating discussion and possible collaboration. Topics may include but are not limited to:

  • Novel probabilistic models for EDAs
  • What features make a suitable application for EDAs?
  • Hybrid probabilistic models for EDAs on mixed variable representations
  • What mathematical tools will advance theoretical study of EDAs?
  • How can theoretical study of EDAs inform the broader EA theory?

Target Participants and Audience

We seek participation from researchers and users of EDAs from the population-based metaheuristics community as well as those interested in deepening the theoretical analysis of this algorithm class.