02/05/2018
Факултетот за електротехника и информациски технологии (ФЕИТ) и Факултетот за информатички науки и компјутерско инженерство (ФИНКИ), заедно со одделот за компјутери и одделот за теорија на информации при македонската секција на IEEE, на 11.05.2018 со почеток во 10:00 часот, во салата за состаноци на ФЕИТ, организираат покането предавање на тема:
,,Deep Statistical Comparison on Meta-heuristic Stochastic Optimization Algorithms“
Предавач: Tome Eftimov, PhD, Computer Systems Department, Jožef Stefan Institute, Ljubljana, Slovenia
Апстракт: To determine the strengths and weaknesses of a selected algorithm, its performance should be compared with performances of state-of-the-art algorithms. The idea behind those comparisons is that by using the results obtained on different problems (e.g., functions, data sets), the "best" algorithm (i.e. algorithm that perform best in average over all problems) can be found, or to use the benchmarking results to transfer the knowledge onto a real-world problem. Statistical analyses that are performed in such cases are crucial and need to be made with a great care because they provide the information from where the conclusions are made, so an appropriate statistical analysis should be performed. Nowadays, many researchers have problems in selecting the right statistic that will be applied on a selected performance measure. Additionally, applying the appropriate statistical test requires knowledge of the necessary conditions about the data that must be met in order to apply it. This kind of misunderstanding is all too common in the research community and can be observed in many high-ranking journal papers. For these reasons, we proposed a novel approach for statistical comparison, known as Deep Statistical Comparison, which provides more robust statistical results than previous state-of-the-art approaches when results are affected by outliers or statistical insignificant difference that could exist between data values. An actual demonstration in which the audience will get familiar with a web-based tool (http://ws.ijs.si/dsc/), designed to make a deep statistical comparison easier, will be given at the end of the talk.
0