Neural Analogical Matching
Authors: Maxwell Crouse, Constantine Nakos, Ibrahim Abdelaziz, Ken Forbus809-817
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | AMN is trained on purely synthetic data and is demonstrated over a diverse set of analogy problems drawn from structure-mapping literature to produce outputs that are largely consistent with SMT. Table 1a shows the results for AMN across different values of r, where r denotes the re-run hyperparameter detailed in Section 4.2. |
| Researcher Affiliation | Collaboration | Maxwell Crouse1 , Constantine Nakos1, Ibrahim Abdelaziz2, Ken Forbus1 1Qualitative Reasoning Group, Northwestern University 2IBM Research, IBM T.J. Watson Research Center |
| Pseudocode | No | The paper describes methods in text and refers to equations in the appendix, but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | The corpus of previous analogy examples was taken from the public release of SME1. (footnote 1http://www.qrg.northwestern.edu/software/sme4/index.html) Visual Oddity: this problem setting was initially proposed to explore cultural differences to geometric reasoning in (Dehaene et al. 2006). Moral Decision Making: this domain was taken from (Dehghani et al. 2008a)... Geometric Analogies: this domain is from one of the first computational analogy experiments (Evans 1964). |
| Dataset Splits | No | The paper states 'AMN was trained on 100,000 synthetic analogy examples' and mentions 'this domain consisted of 1000 examples generated with the same parameters as the training data' for evaluation, implying a test set. However, no specific train/validation/test splits or percentages for validation are explicitly provided. |
| Hardware Specification | No | The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions several techniques and models (e.g., Transformers, DAG LSTMs, Adam) but does not provide specific version numbers for any software dependencies (e.g., Python, PyTorch, TensorFlow, specific library versions). |
| Experiment Setup | No | The paper states 'with the hyperparameters used for AMN provided in Appendix 7.1 (in the supplementary material),' indicating the details are not in the main text. |