Polyhedron Attention Module: Learning Adaptive-order Interactions
Authors: Tan Zhu, Fei Dou, Xinyu Wang, Jin Lu, Jinbo Bi
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results demonstrate the superior classification performance of PAM on massive datasets of the click-through rate prediction and PAM can learn meaningful interaction effects in a medical problem. We evaluate the effectiveness and efficiency of PAM on three large-scale datasets: the Criteo1 and Avazu1 click-through-rate (CTR) datasets, and the UK Biobank2 medical database. We conduct an analysis of the hyperparameters of PAM, and perform ablation studies by individually removing each of the three key components of PAM and evaluating the performance variations. |
| Researcher Affiliation | Academia | Tan Zhu University of Connecticut tan.zhu@uconn.edu Fei Dou University of Georgia fei.dou@uga.edu Xinyu Wang University of Connecticut xinyu.wang@uconn.edu Jin Lu University of Georgia jin.lu@uga.edu Jinbo Bi University of Connecticut jinbo.bi@uconn.edu |
| Pseudocode | Yes | Algorithm 1: Obtain ϕI for an input x |
| Open Source Code | No | The paper does not include a specific link to the source code for the proposed Polyhedron Attention Module (PAM) or an explicit statement confirming its release. |
| Open Datasets | Yes | Both the Criteo and the Avazu are massive industry datasets containing feature values and click feedback for display ads, and are processed following the benchmark protocol in BARS [35, 36]. The UK Biobank serves as a comprehensive biomedical database and research resource, offering extensive genetic and health-related information, where our objective is to predict participants age by leveraging the grey matter volumes from 139 distinct brain regions. Footnotes: 1https://github.com/openbenchmark/BARS/tree/master; 2https://www.ukbiobank.ac.uk/ |
| Dataset Splits | Yes | Table 1: Statistics of the datasets. Dataset #Train #Valid #Test #Features Criteo 33M 8.3M 4.6M 2.1M Avazu 28.3M 4M 8.1M 1.5M UK Biobank 31.8K 4K 4K 139 |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., exact GPU/CPU models, memory, or cloud instance types) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific software dependency details, such as library names with version numbers (e.g., 'PyTorch 1.9', 'Python 3.8'), needed to replicate the experiment. |
| Experiment Setup | No | The paper discusses a 'Hyper-parameter Study' and mentions testing initial values for Ui and the effects of tree depth D, but it does not provide specific hyperparameter values (e.g., learning rate, batch size, optimizer settings) or other concrete training configurations for the final reported models within the provided text. It refers to 'Appendix G' for implementation details, which is not provided. |