Inferning with High Girth Graphical Models

Authors: Uri Heinemann, Amir Globerson

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

Reproducibility Variable Result LLM Response
Research Type Experimental Results on synthetic data show that the models we learn indeed outperform those obtained by other algorithms, which do not return high girth graphs.
Researcher Affiliation Academia Uri Heinemann URIHEI@CS.HUJI.AC.IL The Hebrew University of Jerusalem, Jerusalem, Israel Amir Globerson GAMIR@CS.HUJI.AC.IL The Hebrew University of Jerusalem, Jerusalem, Israel
Pseudocode Yes Algorithm 1 Extended Chow Liu
Open Source Code No The paper does not provide an explicit statement about releasing its source code or a link to a code repository for the methodology described.
Open Datasets No The paper uses 'synthetic data' that was generated by the authors, with details on how it was generated ('starting with a random tree structure and then adding random edges', 'parameters hi were drawn from a uniform distribution', 'parameters Jij were drawn from a uniform distribution'), but does not provide access information (link, citation, etc.) to this data.
Dataset Splits No The paper refers to a 'training sample' and evaluates models based on '100 random queries' from synthetic data, but it does not specify explicit train/validation/test dataset splits with percentages, sample counts, or specific predefined split methodologies.
Hardware Specification No The paper does not provide any specific details regarding the hardware used for running the experiments, such as GPU or CPU models, or cloud computing specifications.
Software Dependencies No The paper describes algorithms like 'loopy belief propagation' and 'Chow Liu algorithm' but does not specify any software dependencies with version numbers used for the implementation or experiments.
Experiment Setup Yes All the models considered have p = 20 variables, so as to allow exact inference for comparisons. The underlying graphs were constrained to have a girth of g = 8. [...] The field parameters hi were drawn from a uniform distribution on [ 0.1, 0.1]. The scale of the interaction parameters Jij varied, as described next. [...] The parameters Jij were drawn from a uniform distribution on [ 1.1, 1.1]. [...] the number of samples is always n = 3200.