Exchangeable Neural ODE for Set Modeling

Authors: Yang Li, Haidong Yi, Christopher Bender, Siyuan Shan, Junier B. Oliva

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the efficacy of our method over strong baselines. ... 4 Experiment The experiments are divided into three parts. First, we evaluate Ex NODE on point cloud classification (Sec. 4.2). Second, we conduct experiments to validate the efficacy of Ex NODE for point cloud generation and likelihood estimation (Sec. 4.3). Finally, we explore the temporal set modeling task (Sec. 4.4)... Table 1: Test Accuracy for point cloud classification... Table 2: Per Point Log Likelihood (PPLL) on test set.
Researcher Affiliation Academia Yang Li Department of Computer Science University of North Carolina at Chapel Hill yangli95@cs.unc.edu Haidong Yi Department of Computer Science University of North Carolina at Chapel Hill haidyi@cs.unc.edu Christopher M. Bender Department of Computer Science University of North Carolina at Chapel Hill bender@cs.unc.edu Siyuan Shan Department of Computer Science University of North Carolina at Chapel Hill siyuanshan@cs.unc.edu Junier B. Oliva Department of Computer Science University of North Carolina at Chapel Hill joliva@cs.unc.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes We post our code at https://github.com/lupalab/Ex NODE.
Open Datasets Yes We evaluate Ex NODE on point cloud classification using Model Net40 [26]... we conduct experiments for set generative task using Spatial MNIST [3] and Model Net40 [3].
Dataset Splits No The paper mentions training and testing but does not provide specific details about a validation dataset split.
Hardware Specification Yes The generative flow models could take roughly 4 days on one TITAN XP GPU
Software Dependencies No Our implementation of neural ODE utilizes the official implementation of the NODE [6]. We use dopri5 for flow models and RK4 for classification models.
Experiment Setup No Architectural details are provided in Appendix D.