LBF

LBF copied to clipboard

This repository is the implementation for the AAAI2021 accepted paper:

Yining Hong, Qing Li, Daniel Ciao, Siyuan Huang and Song-Chun Zhu Learning by Fixing: Solving Math Word Problems with Weak Supervision AAAI2021.

Checkpoint

Checkpoint is provided here. The checkpoint can achieve 59.8% accuracy, which is slightly better than the result reported in the paper.

Seq2Tree Model

A Seq2Tree Neural Network containing top-down Recursive Neural Network and bottom-up Recursive Neural Network

Learning by Fixing

An abductive learning framework that can fix the wrong expressions generated by Seq2Tree

Requirements

• python 3
• PyTorch 0.4.1

Train and Test

• Math23K:

python run_seq2tree.py --model ma-fix --nstep 50 --name ma-fix
python run_seq2tree.py --model fix --nstep 50 --name fix
python run_seq2tree.py --model reinforce --name reinforce
python run_seq2tree.py --model mapo --name mapo 

Citation

@inproceedings{hong2021weakly,
  title     = {Learning by Fixing: Solving Math Word Problems with Weak Supervision},
  author    = {Hong, Yining and Li, Qing and Ciao, Daniel and Huang, Siyuan and Zhu, Song-Chun},
  booktitle = {Proceedings of the Thirty-Fifth AAAI Conference on Artificial Intelligence, {AAAI-21}},            
  year      = {2021}
}
@inproceedings{Li2020ClosedLN,
  title={Closed Loop Neural-Symbolic Learning via Integrating Neural Perception, Grammar Parsing, and Symbolic Reasoning},
  author={Qing Li and Siyuan Huang and Yining Hong and Y. Chen and Y. Wu and S. Zhu},
  journal={Proceedings of the Thirty-eighth International Conference on Machine Learning},
  year={2020}
}