In this lecture (which unfortunately does not have a video captured due to technical problems in the room), we cover several different niching methods that evolutionary (population-based) multi-modal optimization techniques use to stabilize populations across all of the different peaks of an optimization function. We motivate why peak finding is a potentially useful optimization objective and describe how each approach (which are primarily designed to integrate into genetic algorithms but could be used in other evolutionary algorithms) differs not only in its performance but its computational cost. Ultimately, we would like to distribute very many individuals across the many peaks, which means that we seek algorithms that scale manageably as we scale up the number of individuals. By the end of the lecture, restricted tournament selection (RTS) is discussed as being very useful in multi-modal optimization both in terms of its good performance and hyperparameters that allow for trading off speed and accuracy.
Whiteboard notes for this lecture can be found at:
No comments:
Post a Comment