POB2 seminar - "Instance selection methods - past, present and future"
On 26 May 2022 at 3 p.m. the next scientific seminar of Priority Research Area 2 - Artificial Intelligence and Data Processing will take place. The meeting is scheduled in remote mode.
During the seminar, participants will have the opportunity to attend a lecture by Dr. Álvar Arnaiz from the University of Burgos, Spain, who will give a presentation "Instance selection methods - past, present and future".
The event will start at 3 p.m. and will be conducted in English.
Instance selection attempts to find the most representative subset, of the initial training data set, without lessening the predictive capabilities of original one. In other words, if we train one algorithm with the original data set, and another with the selected subset, both algorithms must perform in a similar manner. Instance selection can be seen as a special case of instance generation, where the instances to be generated are limited to the original ones. These methods play a central role in data reduction processes. When real-world data sets are examined, the imperative need for instance selection algorithms becomes increasingly clear. On the one hand, the average data set size is becoming larger and larger. On the other hand, real data sets usually contain noisy instances, outliers, and anomalies.
The first instance selection methods emerged half a century ago, since then several algorithms have been presented: for classification, for regression and for a range of datasets. The presentation will be focused on instance selection methods, their beginning, their achievements along the time and the future research lines that are still opened.
