Neuro-Symbolic Hierarchical Rule Induction

Authors: Claire Glanois, Zhaohui Jiang, Xuening Feng, Paul Weng, Matthieu Zimmer, Dong Li, Wulong Liu, Jianye Hao

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically validate our model on various tasks (ILP, visual genome, reinforcement learning) against relevant state-of-the-art methods, including traditional ILP methods and neurosymbolic models. Our contributions can be summarized as follows: (4) Empirical validation on various domains (see Section 6).
Researcher Affiliation Collaboration 1IT University of Copenhagen, Denmark 2UM-SJTU Joint Institute, Shanghai Jiao Tong University, Shanghai, China 3Huawei Noah s Ark Lab, China 4School of Computing and Intelligence, Tianjin University.
Pseudocode No The paper describes the inference steps using mathematical equations but does not provide any explicit pseudocode blocks or algorithm figures.
Open Source Code Yes The source code of HRI and the scripts to reproduce the experimental results can be found at <https://github.com/claireaoi/hierarchical-rule-induction>.
Open Datasets Yes For (3), we consider a large domain from Visual Genome (Krishna et al., 2017)). Our model outperforms other methods such as NLIL (Yang & Song, 2020) on those tasks. We also empirically validate all our design choices. GQA (Hudson & Manning, 2019b) which is a preprocessed version of the Visual Genome dataset (Krishna et al., 2017)
Dataset Splits Yes We use randomly generated training data for this task given the range of integers. Hyperparameter details are given in Appendix E. In Table 14, the hyperparameters train-num-constants and eval-num-constants represent the number of constants during training and evaluation, respectively.
Hardware Specification No The paper mentions that its implementation uses GPU and is 'GPU-based', but does not specify any particular GPU model, CPU, or other hardware components used for the experiments.
Software Dependencies No The paper does not explicitly state specific software dependencies with version numbers (e.g., Python, PyTorch, or other libraries).
Experiment Setup Yes Hyperparameter details are given in Appendix E. We list relevant generic and task-specific hyperparameters used for our training method in Tables 13 and 14, respectively. For instance, Table 13 lists 'temperature', 'Gumbel-Scale', 'lr', and Table 14 lists 'max-depth', 'train-steps', 'eval-steps', 'train-num-constants'.