Date: | 2022, June 21 |
Time: | 10:45 a. m. |
Place: | G29-018 |
Author: | Mossakowski, Till |
Title: | Neural Probabilistic Logic Programming in DeepProbLog |
Last week, Janna and Martin have presented their ESC-Rules. This motivates a look at other rule-based approaches that integrate weights. ProbLog is probabilistic Prolog. Weights can be associated to facts and rules, and also to the resulting inferences. This has a certain similarity to Bayes' networks and allows for the more realistic modeling of real-world situations, where often a classical two-valued logic is too simple. DeepProbLog extends ProbLog in that the weights are not manually-designed, but rather learned using a neural network.