Relational Program Synthesis with Numerical Reasoning

Authors: Céline Hocquette, Andrew Cropper

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on four diverse domains, including game playing and program synthesis, show that our approach can (i) learn programs with numerical values from linear arithmetical reasoning, and (ii) outperform existing approaches in terms of predictive accuracies and learning times.
Researcher Affiliation Academia University of Oxford celine.hocquette@cs.ox.ac.uk, andrew.cropper@cs.ox.ac.uk
Pseudocode No The paper describes the steps of its approach, such as program search and numerical search, but it does not present these steps in a formal pseudocode block or an explicitly labeled algorithm figure.
Open Source Code Yes The experimental code and data are available at https://github.com/ celinehocquette/numsynth-aaai23.
Open Datasets Yes The experimental code and data are available at https://github.com/ celinehocquette/numsynth-aaai23.
Dataset Splits No The paper does not explicitly provide training/validation/test splits (e.g., percentages or counts) or refer to a specific validation set. It focuses on training and testing.
Hardware Specification Yes We use an 8-Core 3.2 GHz Apple M1 and a single CPU.
Software Dependencies No The paper mentions POPPER (with version 2.0.0 in a footnote), Z3 (with a citation year 2008), Clingo (with a citation year 2014), and Prolog, but it does not provide specific version numbers for multiple key software components to fully replicate the environment.
Experiment Setup No The paper mentions a timeout of 10 minutes per task, measuring mean and standard error over 10 trials, and the hardware used. However, it does not specify concrete hyperparameters (e.g., learning rate, batch size, number of epochs) or detailed system-level training configurations beyond the hardware and time limit.