An Instability in Variational Inference for Topic Models

Authors: Behrooz Ghorbani, Hamid Javadi, Andrea Montanari

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We carried out extensive numerical simulations with the naive mean field iteration. ... In Figure 1 we plot empirical results for the average V(c W ), V(c H) for k = 2, ν = 1 and four values of δ, within the Gaussian model. In Figure 2 (left frame), we plot the empirical probability that variational inference does not converge to the uninformative fixed point or, more precisely, b P(V(c W ) ε0) with ε0 = 10 4, evaluated on a grid of (β, δ) values, for the same model.
Researcher Affiliation Academia 1Department of Electrical Engineering, Stanford University, CA 2Digital Signal Processing Group, Rice University, TX 3Department of Statistics, Stanford University, CA.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks (clearly labeled algorithm sections or code-like formatted procedures).
Open Source Code No The paper does not include an unambiguous statement that the authors are releasing the source code for the work described, nor does it provide a direct link to a source-code repository.
Open Datasets No We select a two-dimensional grid of (δ, β) s and generate 400 different instances according to the LDA model for each grid point. The paper uses synthetic data generated according to a model, not a publicly available dataset, and thus provides no concrete access information to a public dataset.
Dataset Splits No The paper does not explicitly provide training/test/validation dataset splits (e.g., exact percentages, sample counts, or citations to predefined splits) needed to reproduce the data partitioning. It uses synthetic data generated from a model for '400 random realizations'.
Hardware Specification No The paper mentions 'numerical simulations' but does not provide any specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes We initialize the naive mean field iteration near the uninformative fixed-point and iterate until a convergence criterion or the maximum number of 300 iterations is reached.