Peter Korošec

Journal articlesshow all

Using Machine Learning Methods to Assess Module Performance Contribution in Modular Optimization Frameworks, ECJ (2024)
Opt2Vec - A continuous optimization problem representation based on the algorithm: a case study on problem classification, INFORM SCIENCES (2024)

Conference papersshow all

Per-Run Algorithm Performance Improvement Forecasting Using Exploratory Landscape Analysis Features: A Case Study in Single-Objective Black-Box Optimization, GECCO 2024
Analyzing the Generalizability of Automated Algorithm Selection: A Case Study for Numerical Optimization, IEEE SSCI 2023
PS-AAS: Portfolio Selection for Automated Algorithm Selection in Black-Box Optimization, AutoML 2023
Sensitivity Analysis of RF+clust for Leave-one-problem-out Performance Prediction, IEEE CEC 2023
DynamoRep: Trajectory-Based Population Dynamics for Classification of Black-box Optimization Problems, GECCO 2023
Assessing the Generalizability of a Performance Predictive Model, GECCO 2023
Algorithm Instance Footprint: Separating Easily Solvable and Challenging Problem Instances, GECCO 2023

Books & Chaptersshow all

Reliability Improvements for In-Wheel Motor, Recent Advances in Microelectronics Reliability, Springer (2024)
Deep Statistical Comparison for Meta-heuristic Stochastic Optimization Algorithms, Deep Statistical Comparison for Meta-heuristic Stochastic Optimization Algorithms, Springer Cham (2022)
Multilevel optimisation, Optimization Under Uncertainty with Applications to Aerospace Engineering, Springer (2021)

Other publicationsshow all

A Novel Approach to Statistical Comparison of Meta-heuristic Stochastic Optimization Algorithms using Deep Statistics

T. Eftimov, P. Korošec, B. Koroušić Seljak
DOI: 10.1016/j.ins.2017.07.015

BibTex:
@article{Eftimov2017186, title = "A Novel Approach to statistical comparison of meta-heuristic stochastic optimization algorithms using deep statistics ", journal = "Information Sciences ", volume = "417", number = "", pages = "186 - 215", year = "2017", note = "", issn = "0020-0255", doi = "https://doi.org/10.1016/j.ins.2017.07.015", url = "http://www.sciencedirect.com/science/article/pii/S002002551631194X", author = "Tome Eftimov and Peter Koro\v{s}ec and Barbara Korou\v{s}ić\'{c} Seljak", keywords = "Statistical comparison", keywords = "Stochastic optimization algorithms", keywords = "Single objective function", keywords = "Ranking scheme", keywords = "Multiple problem analysis", abstract = "Abstract In this paper a novel approach for making a statistical comparison of meta-heuristic stochastic optimization algorithms over multiple single-objective problems is introduced, where a new ranking scheme is proposed to obtain data for multiple problems. The main contribution of this approach is that the ranking scheme is based on the whole distribution, instead of using only one statistic to describe the distribution, such as average or median. Averages are sensitive to outliers (i.e., the poor runs of the stochastic optimization algorithms) and consequently medians are sometimes used. However, using the common approach with either averages or medians, the results can be affected by the ranking scheme that is used by some standard statistical tests. This happens when the differences between the averages or medians are in some Ïµ-neighborhood and the algorithms obtain different ranks though they should be ranked equally given the small differences that exist between them. The experimental results obtained on Black-Box Benchmarking 2015, show that our approach gives more robust results compared to the common approach in cases when the results are affected by outliers or by a misleading ranking scheme. " }

abstract:
In this paper a novel approach for making a statistical comparison of meta-heuristic stochastic optimization algorithms over multiple single-objective problems is introduced, where a new ranking scheme is proposed to obtain data for multiple problems. The main contribution of this approach is that the ranking scheme is based on the whole distribution, instead of using only one statistic to describe the distribution, such as average or median. Averages are sensitive to outliers (i.e. the poor runs of the stochastic optimization algorithms) and consequently medians are sometimes used. However, using the common approach with either averages or medians, the results can be affected by the ranking scheme that is used by some standard statistical tests. This happens when the differences between the averages or medians are in some ϵ-neighborhood and the algorithms obtain different ranks though they should be ranked equally given the small differences that exist between them. The experimental results obtained on Black-Box Benchmarking 2015, show that our approach gives more robust results compared to the common approach in cases when the results are affected by outliers or by a misleading ranking scheme.