Clamping Variables and Approximate Inference

Authors: Adrian Weller, Tony Jebara

NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present experiments in 5, demonstrating that clamping even a single variable selected using a simple heuristic can be very beneficial. and Results are displayed in Figures 2 to 4 showing average absolute error of log ZB vs log Z and average ℓ1 error of singleton marginals.
Researcher Affiliation Academia Adrian Weller Columbia University, New York, NY 10027 adrian@cs.columbia.edu Tony Jebara Columbia University, New York, NY 10027 jebara@cs.columbia.edu
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any specific links to open-source code or explicitly state that the code for the described methodology is available.
Open Datasets No Test models were constructed as follows: For n variables, singleton potentials were drawn θi U[ Tmax, Tmax]; edge weights were drawn Wij U[0, Wmax] for attractive models, or Wij U[ Wmax, Wmax] for general models. For models with random edges, we constructed Erd os-Renyi random graphs (rejecting disconnected samples). This indicates the authors generated their own data rather than using a publicly available dataset.
Dataset Splits No The paper does not specify training, validation, or test dataset splits. The authors constructed test models by drawing parameters and generating random graphs rather than using pre-existing datasets with established splits.
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments. It mentions 'FW provides no runtime guarantee... but across all parameters, the average combined runtimes on the two clamped submodels was the same order of magnitude as that for the original model' but does not specify any GPU, CPU, or other hardware components.
Software Dependencies No The paper mentions software like Frank-Wolfe (FW) and the junction tree algorithm, but does not provide specific version numbers for any of the software dependencies used in the experiments.
Experiment Setup Yes Test models were constructed as follows: For n variables, singleton potentials were drawn θi U[ Tmax, Tmax]; edge weights were drawn Wij U[0, Wmax] for attractive models, or Wij U[ Wmax, Wmax] for general models. For models with random edges, we constructed Erd os-Renyi random graphs... we examined n = 10, p = 0.5 and n = 50, p = 0.1. Also, in this Section we use the same unbiased reparameterization used by Weller et al. (2014), with E = P i V θixi P (i,j) E Wij 2 [xixj + (1 xi)(1 xj)].