Ambiguity-Aware Optimization for Robust Neuro-Symbolic Learning

• Dayana I. Agudo USP
• Christian F. Barreto USP
• Igor C. Silveira USP
• Denis D. Mauá USP

DOI: https://doi.org/10.5753/eniac.2026.31433

Resumo

Reasoning shortcuts are a well-known phenomenon in neurosymbolic models, where high predictive performance is achieved through semantically incorrect concepts. While existing approaches mitigate reasoning shortcuts, they do not explicitly model the structural properties of the task that give rise to this behavior. This work defines structural task ambiguity as the multiplicity of concept configurations compatible with a symbolic output. Moreover, we propose an ambiguity-aware optimization and the RSR metric to quantify shortcut reliance. Experiments on MNIST tasks show that modeling ambiguity improves conceptual alignment maintaining competitive predictive performance, highlighting its importance for robust neuro-symbolic learning.

Referências

Acharya, K. and Song, H. (2026). A comprehensive review of neuro-symbolic ai for robustness, uncertainty quantification, and intervenability. Arabian Journal for Science and Engineering, 51(1):35–67.

Bortolotti, S., Marconato, E., Carraro, T., Morettin, P., van Krieken, E., Vergari, A., Teso, S., and Passerini, A. (2024). A neuro-symbolic benchmark suite for concept quality and reasoning shortcuts. In Proceedings of the 38th International Conference on Neural Information Processing Systems, Red Hook, NY, USA. Curran Associates Inc.

Colelough, B. C. and Regli, W. (2024). Neuro-symbolic ai in 2024: A systematic review. In Proceedings of the First International Workshop on Logical Foundations of Neuro-Symbolic AI, volume 3819 of CEUR Workshop Proceedings, pages 1–12. CEUR-WS.org.

Geh, R. L., Gonçalves, J., Silveira, I. C., Mauá, D. D., and Cozman, F. G. (2023). dpasp: A comprehensive differentiable probabilistic answer set programming environment for neurosymbolic learning and reasoning.

Geh, R. L., Gonçalves, J., Silveira, I. C., Mauá, D. D., and Cozman, F. G. (2024). dPASP: A Probabilistic Logic Programming Environment For Neurosymbolic Learning and Reasoning. In Proceedings of the 21st International Conference on Principles of Knowledge Representation and Reasoning, pages 731–742.

Huang, J., Li, Z., Chen, B., Samel, K., Naik, M., Song, L., and Si, X. (2021). Scallop: From probabilistic deductive databases to scalable differentiable reasoning. In Ranzato, M., Beygelzimer, A., Dauphin, Y., Liang, P., and Vaughan, J. W., editors, Advances in Neural Information Processing Systems, volume 34, page 25134–25145. Curran Associates, Inc.

LeCun, Y., Cortes, C., and Burges, C. (2010). Mnist handwritten digit database. ATT Labs [Online]. Available: [link], 2.

Li, Z., Huang, J., and Naik, M. (2023). Scallop: A language for neurosymbolic programming. Proceedings of the ACM on Programming Languages, 7(4):1463–1487.

Manhaeve, R., Dumancic, S., Kimmig, A., Demeester, T., and De Raedt, L. (2018). Deepproblog: Neural probabilistic logic programming. In Bengio, S., Wallach, H., Larochelle, H., Grauman, K., Cesa-Bianchi, N., and Garnett, R., editors, Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc.

Manhaeve, R., Dumančić, S., Kimmig, A., Demeester, T., and De Raedt, L. (2021). Neural probabilistic logic programming in deepproblog. Artificial Intelligence, 298:103504.

Marconato, E., Bortolotti, S., van Krieken, E., Vergari, A., Passerini, A., and Teso, S. (2024). BEARS make neuro-symbolic models aware of their reasoning shortcuts. In Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence. JMLR.org.

Marconato, E., Teso, S., Vergari, A., and Passerini, A. (2023). Not all neuro-symbolic concepts are created equal: analysis and mitigation of reasoning shortcuts. In Proceedings of the 37th International Conference on Neural Information Processing Systems, Red Hook, NY, USA. Curran Associates Inc.

Stein, A., Velingker, N., Naik, M., and Wong, E. (2025). Neuro-symbolic programming in the age of foundation models: Pitfalls and opportunities. Proceedings of Machine Learning Research, pages 1–18.

Yang, X.-W., Shao, J.-J., Guo, L.-Z., Zhang, B.-W., Zhou, Z., Jia, L.-H., Dai, W.-Z., and Li, Y.-F. (2025). Neuro-symbolic artificial intelligence: towards improving the reasoning abilities of large language models. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, pages 10770–10778.

Yang, X.-W., Wei, W.-D., Shao, J.-J., Li, Y.-F., and Zhou, Z.-H. (2024). Analysis for abductive learning and neural-symbolic reasoning shortcuts. In Proceedings of the 41st International Conference on Machine Learning, pages 56524–56541. JMLR.org.

Yang, Z., Ishay, A., and Lee, J. (2020). Neurasp: Embracing neural networks into answer set programming. In Bessiere, C., editor, Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, pages 1755–1762.