Probabilistic Generating Circuits

Authors: Honghua Zhang, Brendan Juba, Guy Van Den Broeck

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

Reproducibility Variable Result LLM Response
Research Type Experimental This section evaluates PGCs ability to model real data on density estimation benchmarks.
Researcher Affiliation Academia 1Computer Science Department, University of California Los Angeles, USA 2Computer Science Department, Washington University in St. Louis, Missouri, USA.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain an explicit statement about the release of source code or a link to a code repository for the methodology described.
Open Datasets Yes We evaluate PGCs on two density estimation benchmarks: 1. Twenty Datasets (Van Haaren & Davis, 2012)... 2. Amazon Baby Registries... This benchmark has been commonly used to evaluate DPP learners (Gillenwater et al., 2014; Mariet & Sra, 2015; Gartrell et al., 2019).
Dataset Splits Yes We randomly split each dataset into train (70%), valid (10%) and test (20%) sets.
Hardware Specification No The paper does not specify the hardware (e.g., specific GPU or CPU models) used for running the experiments.
Software Dependencies No We implement Simple PGC in Py Torch and learn the parameters by maximum likelihood estimation (MLE). In particular, we use Adam (Kingma & Ba, 2014) as the optimizing algorithm to minimize the negative log likelihoods given the training sets. No specific versions are mentioned for PyTorch or Adam.
Experiment Setup Yes We use a weighted sum over Det PGCs as our model. [...] The structure of a Simple PGC is also governed by two hyperparameters: the number of Det PGCs in the weighted sum (denoted by C) and the maximum number of variables (i.e. k in Figure 2) allowed in each group (denoted by K). We tune C and K by a grid search over the following ranges: K {1, 2, 5, 7} and C {1, 4, 7, 10, 20}. and We implement Simple PGC in Py Torch and learn the parameters by maximum likelihood estimation (MLE). In particular, we use Adam (Kingma & Ba, 2014) as the optimizing algorithm to minimize the negative log likelihoods given the training sets. Regularization is done by setting the weight decay parameter in Adam.