SIRD: Transformers Assisted Step by Step Symbolic Integration

Vaibhav Sharma, Abhinav Nagpal, Muhammed Fatih Balin

Keywords: Transformers, Symbolic Mathematics, Symbolic Integration, NLP, LLM, SIRD

TL;DR: Dataset and benchmarks for interpretable way of solving symbolic integration using transformers

Abstract: Recently, deep learning has gained popularity in solving statistical or approximate problems. However, working with symbolic data has been challenging for neural networks. Despite this, the natural sciences are making strides in utilizing deep learning for various use cases. In this work, we aim to solve the problem of symbolic integration by using deep learning through integral rule prediction, enabling faster search and better interpretability. We propose a novel symbolic integration rules dataset containing 27 million distinct functions and integration rule pairs. We show that by combining a transformer model trained on this dataset into SymPy's integral_steps function, the number of branches explored during the depth-first-search procedure was reduced by a factor of 3 and successfully solve functions that the original version was unable to handle.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9377