Variable Elimination in the Fourier Domain

Authors: Yexiang Xue, Stefano Ermon, Ronan Le Bras, Carla, Bart Selman

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the significance of this approach by applying it to the variable elimination algorithm. Compared with the traditional bucket representation and other approximate inference algorithms, we obtain significant improvements. [...] 5. Experiments
Researcher Affiliation Academia Yexiang Xue YEXIANG@CS.CORNELL.EDU Cornell University, Ithaca, NY, 14853, USA Stefano Ermon ERMON@CS.STANFORD.EDU Stanford University, Stanford, CA, 94305, USA Ronan Le Bras, Carla P. Gomes, Bart Selman {LEBRAS,GOMES,SELMAN}@CS.CORNELL.EDU Cornell University, Ithaca, NY, 14853, USA
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any link or explicit statement about the availability of its own source code. It mentions using Lib DAI and ACE, which are third-party tools.
Open Datasets Yes We compare our inference algorithms on large benchmarks from the UAI 2010 Approximate Inference Challenge (UAI). [...] Uai 2010 approximate inference challenge. http:// www.cs.huji.ac.il/project/UAI10.
Dataset Splits No The paper mentions using UAI 2010 challenge benchmarks and generating synthetic training images, but it does not specify any train/validation/test splits for these datasets or instances. It describes the total number of training images but not how they are partitioned.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions "Lib DAI (Mooij, 2010)" and "ACE (Darwiche & Marquis, 2002)" but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes For a fair comparison, we fix the size of the messages for both Fourier VE and Mini-bucket to 2^10 = 1, 024. [...] we set the message size for Fourier VE to be 1,048,576 (2^20). Because the complexity of the multiplication step in Fourier VE is quadratic in the number of coefficients, we further shrink the message size to 1,024 (2^10) during multiplication. We allow 1,000,000 steps for burn in and another 1,000,000 steps for sampling in the MCMC approach. [...] damping rate of 0.1 and the maximal number of iterations 1,000,000. [...] We train the model using contrastive divergence (Hinton, 2002), with k = 15 steps of blocked Gibbs updates, on 20, 000 such training images.