Integration-free Training for Spatio-temporal Multimodal Covariate Deep Kernel Point Processes

Authors: YIXUAN ZHANG, Quyu Kong, Feng Zhou

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

Reproducibility Variable Result LLM Response
Research Type Experimental In the experimental section, we mainly analyze the difference between MLE and score matching for DKMPP, the improvement in performance of DKMPP over baseline models, as well as the impact of various hyperparameters.
Researcher Affiliation Collaboration Yixuan Zhang China-Austria Belt and Road Joint Laboratory on Artificial Intelligence and Advanced Manufacturing Hangzhou Dianzi University yixuan.zhang@hdu.edu.cn Quyu Kong Alibaba Group kongquyu.kqy@alibaba-inc.com Feng Zhou Center for Applied Statistics and School of Statistics Renmin University of China feng.zhou@ruc.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not include an unambiguous statement about releasing code for the work described, nor does it provide a direct link to a source-code repository.
Open Datasets Yes Crimes in Vancouver4 This dataset is composed of more than 530 thousand crime records, including all categories of crimes committed in Vancouver. Each crime record contains the time and location (latitude and longitude) of the crime. https://www.kaggle.com/datasets/wosaku/crime-in-vancouver NYC Vehicle Collisions5 The New York City vehicle collision dataset contains about 1.05 million vehicle collision records. Each collision record includes the time and location (latitude and longitude). https://data.cityofnewyork.us/Public-Safety/NYPD-Motor-Vehicle-Collisions/h9gi-nx95 NYC Complaint Data6 This dataset contains over 228 thousand complaint records in New York City. Each record includes the date, time, and location (latitude and longitude) of the complaint. https://data.cityofnewyork.us/Public-Safety/NYPD-Complaint-Data-Current-YTD/5uac-w243
Dataset Splits Yes Each dataset is divided into training, validation and test data using a 50%/40%/10% split ratio based on time.
Hardware Specification No The paper does not provide specific hardware details (such as exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper mentions using 'MLPs with Re LU activation functions' and 'pre-trained Distil BERT [27]' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes Both the kernel mixture weight network f and the non-linear transformation g in the deep kernel are implemented using MLPs with Re LU activation functions. We fix the representative points on a regular grid: 5 representative points evenly spaced on each axis, so there are 53 = 125 representative points in total. For Score-DKMPP+, we use a Gaussian noise ϵ N(0, σ2) with σ2 = 0.01. When we tested the effect of representation points, we fix the network with 2 hidden layers and when we tested the effect of network structure, we fix the number of representative points as 125. we set a batch size of 100