Generalisation through Negation and Predicate Invention
Authors: David M. Cerna, Andrew Cropper
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental results on multiple domains show that our approach can improve predictive accuracies and learning times. ... To evaluate the impact of combining negation and PI, our experiments aim to answer the question: Q1 Can negation and PI improve learning performance? ... To answer Q1, we compare the performance of NOPI against POPPER... To answer Q2, we compare the performance of NOPI with and without these constraints. To answer Q3, we compare NOPI against POPPER, ALEPH, and METAGOLSN. ... To answer Q4, we evaluate NOPI on tasks where negation and PI should be unnecessary. |
| Researcher Affiliation | Academia | 1Czech Academy of Sciences Institute of Computer Science (CAS ICS), Prague, Czechia 2University of Oxford, United Kingdom |
| Pseudocode | No | The paper describes the NOPI algorithm but does not include structured pseudocode or an algorithm block in the main text. It refers to an ASP encoding in Appendix F. |
| Open Source Code | Yes | Experimental code may be found in the following repository: github.com/Ermine516/NOPI |
| Open Datasets | Yes | Basic (B). Non-monotonic learning tasks introduced by Siebers and Schmid (2018) and Purgał, Cerna, and Kaliszyk (2022)... Zendo (Z). Bramley et al. (2018) introduce Zendo tasks... Graphs (G). We use commonly used graph problems (Evans and Grefenstette 2018; Glanois et al. 2022)... |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits (e.g., specific percentages or sample counts) needed to reproduce the experiment. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as CPU/GPU models, processor types, or memory amounts used for running the experiments. |
| Software Dependencies | Yes | POPPER uses an ASP program P... with the ASP system Clingo (Gebser et al. 2019). |
| Experiment Setup | Yes | We use a 300s learning timeout for each task and round accuracies and learning times to integer values. We plot 99% confidence intervals. |