UNIPoint: Universally Approximating Point Processes Intensities

Authors: Alexander Soen, Alexander Mathews, Daniel Grixti-Cheng, Lexing Xie9685-9694

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Evaluations on synthetic and real world datasets show that this simpler representation performs better than Hawkes process variants and more complex neural network-based approaches.
Researcher Affiliation Academia Alexander Soen, Alexander Mathews, Daniel Grixti-Cheng, Lexing Xie The Australian National University alexander.soen@anu.edu.au, alex.mathews@anu.edu.au, a500846@anu.edu.au, lexing.xie@anu.edu.au
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes Reference code is available online1. 1https://github.com/alexandersoen/unipoint
Open Datasets Yes MOOC2. A dataset of student interactions in online courses (Kumar, Zhang, and Leskovec 2019), previously used for evaluating neural point processes (Shchur, Biloˇs, and G unnemann 2020). 2https://github.com/srijankr/jodie/
Dataset Splits Yes We fit models for all synthetic and real world datasets, with a 60 : 20 : 20 train-validation-test split.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No Our models are implemented in Py Torch4. 4https://pytorch.org (Paszke et al. 2017) While PyTorch is mentioned, a specific version number is not provided.
Experiment Setup Yes All UNIPoint models tested employ an RNN with 48 hidden units, a batch size of 64, and are trained using Adam (Kingma and Ba 2014) with L2 weight decay set to 10 5. The validation set is used for early stopping: training halts if the validation loss does not improve by more than 10 4 for 100 successive minibatches.