TMPNN: High-Order Polynomial Regression Based on Taylor Map Factorization
Authors: Andrei Ivanov, Stefan Ailuro
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | By benchmarking on Feynman regression, UCI, Friedman1, and real-life industrial datasets, we demonstrate that the proposed method performs comparably to the state-of-the-art regression methods and outperforms them on specific tasks. |
| Researcher Affiliation | Academia | Andrei Ivanov1, Stefan Ailuro2 1Independent Researcher, Toronto, Canada 2Independent Researcher, Moscow, Russia 05x.andrey@gmail.com, ailuro.sm@gmail.com. The affiliations listed are 'Independent Researcher' which does not clearly align with 'universities or public research institutions' or 'corporations or private-sector labs' as defined for classification, nor do the email domains (gmail.com). However, in the context of research papers, independent researchers are typically categorized within the broader academic sphere if not explicitly corporate. |
| Pseudocode | Yes | An example of the second-order mapping (3) is presented in Listing 1. |
| Open Source Code | Yes | More advanced implementation using Keras and Tensor Flow is provided in https://github.com/andiva/tmpnn. |
| Open Datasets | Yes | on a set of 33 regression UCI open access datasets, the Feynman symbolic regression benchmark with 120 datasets, the Friedman-1 dataset, and the publicly available multitarget dataset from the gas and petrochemical processing. ... 1https://github.com/treforevans/uci datasets to access datasets ... Friedman-1 dataset (Friedman 1991)... UCI Yacht Hydrodynamics dataset (Gerritsma et al. 2013). |
| Dataset Splits | Yes | The evaluation was executed with ten different random splits (75% for training, 25% for testing and metric reporting). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments, only general statements about implementation. |
| Software Dependencies | No | More advanced implementation using Keras and Tensor Flow is provided in https://github.com/andiva/tmpnn. We use the Adamax optimizer (Kingma and Ba 2015)... While software components like Keras, TensorFlow, and Adamax are mentioned, specific version numbers for these dependencies are not provided. |
| Experiment Setup | Yes | We use the Adamax optimizer (Kingma and Ba 2015) with default parameters based on our experiments. ... We do not use optimal hyperparameter search for the proposed model and suppose k = 3, p = 5 as default. ... For the TMPNN, we used k = 2, p = 5 with all features and without feature engineering, selection, and regularization. ... The experiments setup and hyperparameters for the considered models can be found in supplementary code. |