While Foundational Models have benefited from “scaling laws” their returns on robustness, interpretability, and generalizability are diminishing. The neurosymbolic paradigm offers a promising path away from the limitations of pure scaling. By integrating symbols, logic, and knowledge structures, we can guide learning and generation processes to be more data-efficient, consistent, and explainable.

The Neurosymbolic Generative Models Special Track at the 19th International Conference on Neurosymbolic Learning and Reasoning (NeSy 2025) aims to bring together researchers working at the intersection of symbolic reasoning and deep generative modelling. We invite submissions that explore how neurosymbolic principles can enhance generative approaches, including large language models, diffusion models, and other generative architectures. Our goal is to foster the development of systems that do more than just “scale up” — they leverage structured knowledge, reasoning capabilities, and integrated symbolic components for creativity, interpretability, and trustworthiness.

Topics of Interest

Submissions are encouraged from all areas related to generative modelling enhanced by neurosymbolic techniques. Relevant topics include, but are not limited to:

  • Adding symbolic constraints, and ontologies for logically coherent outputs;
  • Combining neural and symbolic reasoning for guided generation;
  • Using symbolic reasoning for verifiable outputs, ensuring coherence with known facts;
  • Infusing Knowledge Graphs and ontologies for more context-aware, and accurate generation;
  • Integrating symbolic abstractions for novel scientific, creative, or mathematical content;
  • Incorporating regulatory and ethical principles into generation for aligned, standard-compliant results;
  • Developing new metrics, methodologies, and datasets to evaluate data efficiency, explainability and robustness of generative models.

Contact

For inquiries related to this special track, please contact the track chairs: Thiviyan Thanapalasingam (Sony AI) - thiviyan.t@gmail.com and Kareem Ahmed (University of California, Irvine) - kareem.yousrii@gmail.com.

Submission

Please submit your paper according to the submission guidelines. In the submission form on OpenReview, please select the “Neurosymbolic Generative Models” special track.