oroikono/sigs-symbolic-pde-corpus

SIGS Grammar Production Corpus

This dataset contains the one-hot grammar-production sequences used to train the Grammar-VAE in SIGS: Neuro-Symbolic AI for Analytical Solutions of Differential Equations (Oikonomou et al., ICML 2026).

Structure

Each row contains:

• —inputs: a float32 tensor shaped [grammar productions, sequence length];
• —labels: the corresponding integer production indices shaped [sequence length].

The deterministic default split uses seed 42 with 70% training, 20% validation, and 10% test examples.

Intended use

• —Training or evaluating grammar-guided symbolic generative models.
• —Reproducing the SIGS Grammar-VAE representation-learning stage.
• —Studying structured representations for symbolic differential equations.

Limitations

The corpus reflects the published SIGS grammar and expression-generation procedure. It should not be treated as representative of all valid mathematical expressions or differential equations. Models trained on it inherit its grammar, maximum-length, and sampling biases.

Citation

@misc{oikonomou2026neurosymbolic,
  title={Neuro-Symbolic AI for Analytical Solutions of Differential Equations},
  author={Oikonomou, Orestis and Lingsch, Levi and Grund, Dana and Mishra, Siddhartha and Kissas, Georgios},
  year={2026},
  eprint={2502.01476},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
}