Differentiable Learning of Logical Rules for Knowledge Base Reasoning
Authors: Fan Yang, Zhilin Yang, William W. Cohen
NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, our method outperforms prior work on multiple knowledge base benchmark datasets, including Freebase and Wiki Movies. |
| Researcher Affiliation | Academia | Fan Yang Zhilin Yang William W. Cohen School of Computer Science Carnegie Mellon University {fanyang1,zhiliny,wcohen}@cs.cmu.edu |
| Pseudocode | Yes | Algorithm 1 Recover logical rules from attention vectors |
| Open Source Code | No | The paper mentions implementations of other methods used for comparison (ISG, Trans E) and where to find their code, but does not state that the code for Neural LP is open-source or provide a link to it. |
| Open Datasets | Yes | We use the datasets WN18 and FB15K, which are introduced in [3]. We also considered a more challenging dataset, FB15KSelected [25]... We use the WIKIMOVIES dataset from Miller et al. [16]. We conduct experiments on two benchmark datasets [12] in statistical relation learning. The first dataset, Unified Medical Language System (UMLS)... The second dataset, Kinship... |
| Dataset Splits | Yes | For all the tasks, the data used in the experiment are divided into three files: facts, train, and test. ... We randomly split the datasets into facts, train, test files as described above with ratio 6:2:1. ... We use the same train/validation/test split as in prior work and augment data files with reversed data tuples... |
| Hardware Specification | No | The paper states 'Our system is implemented in Tensor Flow and can be trained end-to-end using gradient methods,' but does not specify any particular hardware (e.g., GPU/CPU models, memory) used for the experiments. |
| Software Dependencies | No | The paper mentions 'Tensor Flow,' 'long short-term memory [9],' and 'mini-batch ADAM [11]' as software components, but does not provide specific version numbers for any of them. |
| Experiment Setup | Yes | The recurrent neural network used in the neural controller is long short-term memory [9], and the hidden state dimension is 128. The optimization algorithm we use is mini-batch ADAM [11] with batch size 64 and learning rate initially set to 0.001. The maximum number of training epochs is 10, and validation sets are used for early stopping. |