Ljupčo Todorovski: Symbolic machine learning on graphs

Date of publication: 24. 1. 2023
Mathematics and theoretical computing seminar
10:00 - 12:00
Jadranska 21, 3.07

In the previous seminar, we overviewed methods for embedding graph vertices in Euclidean space, which allow for converting a set of vertices into a table for classical machine learning methods. Models learned using these approaches have good predictive accuracy but do not allow direct interpretation of their predictions.

This week, we will overview symbolic approaches to machine learning on graphs, which can be used to learn predictive models directly from the vertices and edges in a graph. The results of symbolic learning are interpretable predictive models that immediately explain their predictions.