Inference in Graphical Models via Semidefinite Programming Hierarchies

Authors: Murat A. Erdogdu, Yash Deshpande, Andrea Montanari

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 4, we present numerical experiments with PSOS by solving problems of size up to 10, 000 within several minutes.
Researcher Affiliation Collaboration Murat A. Erdogdu Microsoft Research erdogdu@cs.toronto.edu Yash Deshpande MIT and Microsoft Research yash@mit.edu Andrea Montanari Stanford University montanari@stanford.edu
Pseudocode Yes Algorithm 1: Partial-SOS, Algorithm 2: CLAP: Confidence Lift And Project
Open Source Code No The paper does not provide any concrete statement or link regarding the public availability of the source code for the described methodology.
Open Datasets No The paper uses generated data for image denoising (100x100 binary images with Bernoulli or blockwise noise) and Ising spin glasses (two-dimensional grids with i.i.d. parameters), but does not provide specific access information (link, DOI, formal citation) to a publicly available dataset for reproduction.
Dataset Splits No The paper does not provide specific dataset split information (e.g., percentages, sample counts, or citations to predefined splits) for training, validation, and testing.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not specify any software dependencies with version numbers required to replicate the experiments.
Experiment Setup Yes In applying Algorithm 1, we add diagonals to the grid (see right plot in Figure 1) in order to satisfy the condition (3.1) with corresponding weight e_d. The model parameter 0 is chosen in each case as to approximately optimize the performances under BP denoising. We use an inertia of 0.5 to help convergence [YFW05], and threshold the marginals at 0.5. Similar to BP-SP, we use an inertia of 0.5. We use plaquettes in the grid (contiguous groups of four vertices) as the largest regions, and apply message passing with inertia 0.1 [WF01]. We find that there is little or no improvement beyond r = 10 (cf. Figure 2).