Bounding the Cost of Search-Based Lifted Inference

Authors: David B. Smith, Vibhav G. Gogate

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We ran our Rao-Blackwellised Importance Sampler on three benchmark SRMs and datasets: (1) The friends, smokers and Asthma MLN and dataset described in [19], (2) The web KB MLN for collective classification and (3) The Protein MLN, in which the task is to infer protein interactions from biological data. All models are available from www.alchemy.cs.washington.edu. [...] Figure 2 shows the sample variance of the estimators as a function of time.
Researcher Affiliation Academia David Smith University of Texas At Dallas 800 W Campbell Rd, Richardson, TX 75080 dbs014200@utdallas.edu Vibhav Gogate University of Texas At Dallas 800 W Campbell Rd, Richardson, TX 75080 vibhav.gogate@utdallas.edu
Pseudocode Yes Algorithm 1 Function eval Node(And) [...] Algorithm 2 Function eval Node(Or) [...] Algorithm 3 Function count Path Leaves [...] Algorithm 4 Function make Rao Function [...] Algorithm 5 Function eval Rao Function
Open Source Code No The paper states 'All models are available from www.alchemy.cs.washington.edu', which refers to the models/datasets used, not the source code for the methodology described in the paper. No other explicit statement or link for the authors' source code is provided.
Open Datasets Yes All models are available from www.alchemy.cs.washington.edu.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, or detailed splitting methodology) for training, validation, or test sets.
Hardware Specification No The paper does not provide 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 mentions software components like 'Markov Logic Network (MLN)' and 'Alchemy', but does not provide specific version numbers for these or any other ancillary software dependencies used in the experiments.
Experiment Setup Yes For each model, we set 10% randomly selected ground atoms as evidence, and designated them to have True value. We then estimated the partition function via our Rao-Blackwellised sampler with complexity bounds t0, 10, 100, 1000u (bound of 0 yields the LIS algorithm). We used the uniform distribution as our proposal. We ran each sampler 50 times and computed the sample variance of the estimates.