Lifted Hinge-Loss Markov Random Fields
Authors: Sriram Srinivasan, Behrouz Babaki, Golnoosh Farnadi, Lise Getoor7975-7983
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate our proposed lifted inference algorithm, LHL-MRF, on various real and synthetic datasets. |
| Researcher Affiliation | Academia | Sriram Srinivasan ssriniv9@ucsc.edu UC Santa Cruz Behrouz Babaki Behrouz.Babaki@polymtl.ca Polytechnique Montreal Golnoosh Farnadi gfarnadi@ucsc.edu UC Santa Cruz Lise Getoor getoor@ucsc.edu UC Santa Cruz |
| Pseudocode | No | The paper describes algorithms but does not contain a dedicated pseudocode block or algorithm listing. |
| Open Source Code | Yes | We implemented our models using the PSL open-source Java library1. We ground the rules using the PSL library and then run inference using our own implementation of ADMM in C++2. Note that PSL removes a large number of trivial symmetries during the grounding process by removing trivially satisfied rules (for further information see (Augustine and Getoor 2018)). Removing these simple symmetries ensures the extra symmetries that are obtained during our approach are non-trivial. We use Saucy3 from the RELOOP library to perform color refinement (Mladenov et al. 2016). Footnote 2: https://github.com/linqs/srinivasan-aaai19 |
| Open Datasets | Yes | -Citeseer: This dataset includes 3312 papers in six categories, and 4591 citation links. The goal is to classify documents in a citation network. The original data comes from Citeseer . The details about the model and data can be found in Bach et al. (2017). -Cora: This dataset includes includes 2708 papers in seven categories, and 5429 citation links. The goal is to classify documents in a citation network. The original data comes from Cora . The details about the model and data can be found in Bach et al. (2017). -Wikidata: The dataset contains 419 families and 1,844 family trees. The goal is to perform entity-resolution on a family graph obtained form wikidata by crawling the site for familial relations. The details about the model and data can be found in Kouki et al. (2017). |
| Dataset Splits | No | We randomly assign a label to users and keep 50% of the labels as evidence and another 50% as unknowns to be inferred. |
| Hardware Specification | Yes | All experiments were run on a machine with 16GB RAM and an i5 processor. |
| Software Dependencies | No | The paper mentions software like 'PSL open-source Java library', 'our own implementation of ADMM in C++', and 'Saucy3 from the RELOOP library', but does not provide specific version numbers for these dependencies. |
| Experiment Setup | No | The paper describes the overall approach and model components but does not provide specific hyperparameters (e.g., learning rate, batch size, number of epochs) or detailed training configurations. |