Explanatory Learning: Beyond Empiricism in Neural Networks

Antonio Norelli, Giorgio Mariani, Luca Moschella, Andrea Santilli, Giambattista Parascandolo, Simone Melzi, Emanuele Rodolà

Keywords: explainability, rationalism, deep learning

Abstract: We introduce Explanatory Learning (EL), an explanation-driven machine learning framework to use existing knowledge buried in symbolic sequences expressed in an unknown language. In EL, the burden of interpreting explanations is not left to humans or human-coded compilers, as done in Program Synthesis. Rather, EL calls for a learned interpreter, built upon existing explanations paired with observations of several phenomena. This interpreter can then be used to make predictions on novel phenomena, and even find an explanation for them. We formulate the EL problem as a simple binary classification task, so that common end-to-end approaches aligned with the dominant empiricist view of machine learning could, in principle, solve it. To these models, we oppose Critical Rationalist Networks (CRNs), which instead embrace a rationalist view on the acquisition of knowledge. CRNs express several desired properties by construction, they are truly explainable, can adjust their processing at test-time for harder inferences, and can offer strong confidence guarantees on their predictions.

One-sentence Summary: We introduce Explanatory Learning (EL), an explanation-driven machine learning framework to use existing knowledge buried in symbolic sequences expressed in an unknown language.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2201.10222/code)