Anytime Recursive Best-First Search for Bounding Marginal MAP

Authors: Radu Marinescu, Akihiro Kishimoto, Adi Botea, Rina Dechter, Alexander Ihler7924-7932

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental An empirical evaluation demonstrates the effectiveness of our proposed approach against current solvers.
Researcher Affiliation Collaboration 1IBM Research 2University of California, Irvine
Pseudocode Yes Algorithm 1 describes the node generation and node values propagation during search, while Algorithms 2 and 3 show how the best child node and, respectively, an unsolved child node is selected for expansion.
Open Source Code No The paper does not provide any specific link to source code or an explicit statement about its availability (e.g., 'code available at', 'we release our code').
Open Datasets Yes Our benchmark set includes 3 standard problem domains from grid networks (grid), genetic linkage analysis (pedigree), and medical diagnosis expert systems (promedas). Since the original problems are pure MAP tasks, we generated 5 synthetic MMAP instances for each pure MAP instance by randomly selecting 50% of the variables as MAP variables as suggested previously in (Lee et al. 2016; Marinescu et al. 2017). Therefore, we created 75 grid, 110 pedigree, and 50 promedas MMAP instances. Table 1 shows the typical ranges of the problem instance parameters where n is the number of variables, k is the domain size, w c is the constrained induced width and w s is the induced width of the conditioned summation subproblem (we also include in parenthesis the average values of the latter two measures).
Dataset Splits No The paper describes experiments on 'benchmark problems' and 'instances' but does not specify explicit train/validation/test dataset splits or cross-validation methods for these instances.
Hardware Specification No The paper mentions '20GB of RAM memory limit' and 'time limit was set to 1 hour' but does not provide specific details on CPU, GPU models, or other hardware specifications used for experiments.
Software Dependencies No The paper refers to using 'WMBu(i)' and 'WMBl(10)' schemes but does not list specific software packages or libraries with version numbers required for reproducibility.
Experiment Setup Yes For RBFAOO+ and RBFAOOwe set the overestimation parameter δ to 2.0 and their garbage collection scheme replaced R = 30% of the cache entries. The cutoff parameter θ used by LAOBF to trigger the depth-first lookaheads was set to 1000, while Any SBFS and Any LDFS were run with the default parameters specified in (Marinescu, Dechter, and Ihler 2018).