Summary

We started this chapter by discussing the thought process behind developing an NSAI solution based on the powerful technique of LTNs. Then, we saw how LTNs combine symbolic rules and NNs to learn the relationships between the different logical conditions and their variables (i.e., the knowledge base). Thanks to the powerful LTNtorch package, we demonstrated how to quickly build an LTN system for a binary classification task. For example, we used the publicly available Red and White Wine Dataset. We showed the power and benefits of NSAI (specifically, LTNs) in training performance. Our LTN system reached high predictive power much quicker when compared to other public experiments on the same dataset.

In this chapter, we also discussed a more straightforward approach to NSAI. In this example, we directly combined a DT (as the symbolic AI component) and an NN as a stacking classifier. We showed how we can exploit the logic extraction capabilities of the DT to further understand...