Interpretable Neural-Symbolic Concept Reasoning

Pietro Barbiero, Gabriele Ciravegna, Francesco Giannini, Mateo Espinosa Zarlenga, Lucie Charlotte Magister, Alberto Tonda, Pietro Lio, Frederic Precioso, Mateja Jamnik, Giuseppe Marra
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:1801-1825, 2023.

Abstract

Deep learning methods are highly accurate, yet their opaque decision process prevents them from earning full human trust. Concept-based models aim to address this issue by learning tasks based on a set of human-understandable concepts. However, state-of-the-art concept-based models rely on high-dimensional concept embedding representations which lack a clear semantic meaning, thus questioning the interpretability of their decision process. To overcome this limitation, we propose the Deep Concept Reasoner (DCR), the first interpretable concept-based model that builds upon concept embeddings. In DCR, neural networks do not make task predictions directly, but they build syntactic rule structures using concept embeddings. DCR then executes these rules on meaningful concept truth degrees to provide a final interpretable and semantically-consistent prediction in a differentiable manner. Our experiments show that DCR: (i) improves up to +25% w.r.t. state-of-the-art interpretable concept-based models on challenging benchmarks (ii) discovers meaningful logic rules matching known ground truths even in the absence of concept supervision during training, and (iii), facilitates the generation of counterfactual examples providing the learnt rules as guidance.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-barbiero23a, title = {Interpretable Neural-Symbolic Concept Reasoning}, author = {Barbiero, Pietro and Ciravegna, Gabriele and Giannini, Francesco and Espinosa Zarlenga, Mateo and Magister, Lucie Charlotte and Tonda, Alberto and Lio, Pietro and Precioso, Frederic and Jamnik, Mateja and Marra, Giuseppe}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {1801--1825}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/barbiero23a/barbiero23a.pdf}, url = {https://proceedings.mlr.press/v202/barbiero23a.html}, abstract = {Deep learning methods are highly accurate, yet their opaque decision process prevents them from earning full human trust. Concept-based models aim to address this issue by learning tasks based on a set of human-understandable concepts. However, state-of-the-art concept-based models rely on high-dimensional concept embedding representations which lack a clear semantic meaning, thus questioning the interpretability of their decision process. To overcome this limitation, we propose the Deep Concept Reasoner (DCR), the first interpretable concept-based model that builds upon concept embeddings. In DCR, neural networks do not make task predictions directly, but they build syntactic rule structures using concept embeddings. DCR then executes these rules on meaningful concept truth degrees to provide a final interpretable and semantically-consistent prediction in a differentiable manner. Our experiments show that DCR: (i) improves up to +25% w.r.t. state-of-the-art interpretable concept-based models on challenging benchmarks (ii) discovers meaningful logic rules matching known ground truths even in the absence of concept supervision during training, and (iii), facilitates the generation of counterfactual examples providing the learnt rules as guidance.} }
Endnote
%0 Conference Paper %T Interpretable Neural-Symbolic Concept Reasoning %A Pietro Barbiero %A Gabriele Ciravegna %A Francesco Giannini %A Mateo Espinosa Zarlenga %A Lucie Charlotte Magister %A Alberto Tonda %A Pietro Lio %A Frederic Precioso %A Mateja Jamnik %A Giuseppe Marra %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-barbiero23a %I PMLR %P 1801--1825 %U https://proceedings.mlr.press/v202/barbiero23a.html %V 202 %X Deep learning methods are highly accurate, yet their opaque decision process prevents them from earning full human trust. Concept-based models aim to address this issue by learning tasks based on a set of human-understandable concepts. However, state-of-the-art concept-based models rely on high-dimensional concept embedding representations which lack a clear semantic meaning, thus questioning the interpretability of their decision process. To overcome this limitation, we propose the Deep Concept Reasoner (DCR), the first interpretable concept-based model that builds upon concept embeddings. In DCR, neural networks do not make task predictions directly, but they build syntactic rule structures using concept embeddings. DCR then executes these rules on meaningful concept truth degrees to provide a final interpretable and semantically-consistent prediction in a differentiable manner. Our experiments show that DCR: (i) improves up to +25% w.r.t. state-of-the-art interpretable concept-based models on challenging benchmarks (ii) discovers meaningful logic rules matching known ground truths even in the absence of concept supervision during training, and (iii), facilitates the generation of counterfactual examples providing the learnt rules as guidance.
APA
Barbiero, P., Ciravegna, G., Giannini, F., Espinosa Zarlenga, M., Magister, L.C., Tonda, A., Lio, P., Precioso, F., Jamnik, M. & Marra, G.. (2023). Interpretable Neural-Symbolic Concept Reasoning. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:1801-1825 Available from https://proceedings.mlr.press/v202/barbiero23a.html.

Related Material