Neural Sequence-to-grid Module for Learning Symbolic Rules
Authors: Segwang Kim, Hyoungwook Nam, Joonyoung Kim, Kyomin Jung8163-8171
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that neural networks having our module as an input preprocessor achieve OOD generalization on various arithmetic and algorithmic problems including number sequence prediction problems, algebraic word problems, and computer program evaluation problems while other state-of-the-art sequence transduction models cannot. |
| Researcher Affiliation | Academia | 1 Department of Electrical and Computer Engineering, Seoul National University, Seoul, Republic of Korea 2 Department of Computer Science, University of Illinois at Urbana-Champaign, Urbana, Illinois, USA |
| Pseudocode | No | The paper describes the nested list operations mathematically and provides formulas, but it does not include a structured pseudocode block or a clearly labeled algorithm figure. |
| Open Source Code | Yes | Our implementations3 based on the open source library tensor2tensor4 contain detailed training schemes and hyperparameters of our models and the baselines. 3https://github.com/Segwang Kim/neural-seq2grid-module |
| Open Datasets | Yes | Number Sequence Prediction As the name suggests, the goal of the number sequence prediction problem (Nam, Kim, and Jung 2019) is to predict the next term of an integer sequence. |
| Dataset Splits | No | The paper mentions training and two test sets (ID and OOD) for the main arithmetic and algorithmic problems, with specific sizes (1M training, 10K for each test set). For a toy problem, it mentions validating on six separate validation sets. However, it does not provide specific details on how validation sets were created or used for the main experiments. |
| Hardware Specification | Yes | All models could fit in a single NVIDIA GTX 1080ti GPU. |
| Software Dependencies | No | The paper states 'Our implementations3 based on the open source library tensor2tensor4 contain detailed training schemes and hyperparameters of our models and the baselines.', but does not provide specific version numbers for software components or libraries. |
| Experiment Setup | Yes | We determined configurations of our models by hyperparameter sweeping for each problem. |