Dependent Multinomial Models Made Easy: Stick-Breaking with the Polya-gamma Augmentation

Authors: Scott Linderman, Matthew J. Johnson, Ryan P. Adams

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental After providing background material and defining our general models and inference methods, we demonstrate the utility of this class of models by applying it to three domains as case studies. First, we develop a correlated topic model for text corpora. Second, we study an application to modeling the spatial and temporal patterns in birth names given only sparse data. Finally, we provide a new continuous state-space model for discrete-valued sequences, including text and human DNA.
Researcher Affiliation Collaboration Scott W. Linderman Harvard University Cambridge, MA 02138 swl@seas.harvard.edu Matthew J. Johnson Harvard University Cambridge, MA 02138 mattjj@csail.mit.edu Ryan P. Adams Twitter & Harvard University Cambridge, MA 02138 rpa@seas.harvard.edu
Pseudocode No The paper describes algorithms but does not include any explicit pseudocode blocks or sections labeled 'Algorithm'.
Open Source Code Yes Code to use these models, write new models that leverage these inference methods, and reproduce the figures in this paper is available at github.com/HIPS/pgmult.
Open Datasets Yes Figure 2 shows results on both the AP News dataset and the 20 Newsgroups dataset, where models were trained on a random subset of 95% of the complete documents and tested on the remaining 5% by estimating held-out likelihoods of half the words given the other half.
Dataset Splits No models were trained on a random subset of 95% of the complete documents and tested on the remaining 5% by estimating held-out likelihoods of half the words given the other half.
Hardware Specification No The paper does not provide any specific hardware details used for running its experiments.
Software Dependencies No The paper mentions using 'off-the-shelf algorithms and software for Gaussian processes and linear Gaussian dynamical systems' but does not provide specific software names with version numbers.
Experiment Setup No The paper describes model architectures and inference algorithms, such as a '10-dimensional state space' for linear dynamical models and '10 discrete states' for HMMs, but it does not provide specific hyperparameter values or detailed training configurations like learning rates, batch sizes, or optimizer settings.