Uprooting and Rerooting Graphical Models

Authors: Adrian Weller

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide an empirical evaluation in 6, showing that rerooting can be particularly effective for models with dense, strong edges and weak singleton potentials. Section 6. Experiments: We ran experiments on the following topologies and model sizes: complete graphs on 10 and 15 variables; grids of size 5 x 5 and 9 x 9. All potentials were drawn randomly.
Researcher Affiliation Academia Adrian Weller ADRIAN.WELLER@ENG.CAM.AC.UK Department of Engineering, University of Cambridge, United Kingdom
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code No The paper states 'All methods were implemented using lib DAI (Mooij, 2010)' which is a third-party library, but does not provide access to the authors' own source code for the methodology described in the paper.
Open Datasets No The paper evaluates inference methods on graphical models with specified topologies (complete graphs, grids) where potentials were drawn randomly. It does not use a fixed, publicly available dataset with a training set in the typical sense.
Dataset Splits No The paper evaluates inference methods on graphical models with specified topologies and random potentials. It does not refer to traditional train/validation/test splits as it is not a supervised learning task on a fixed dataset.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments (e.g., GPU models, CPU types, memory).
Software Dependencies Yes All methods were implemented using lib DAI (Mooij, 2010), see the Appendix 9 for details.
Experiment Setup Yes All potentials were drawn randomly: mixed models used Wij U[ Wmax, Wmax], attractive models used Wij U[0, Wmax], as Wmax was varied; singleton potentials were drawn either from a low range θi [ 0.1, 0.1], medium range θi [ 2, 2], or from a range commensurate with edge potentials, i.e. θi U[ Wmax/2, Wmax/2], with the factor of 2 needed given the form of (1).