Factored Probabilistic Belief Tracking

Authors: Blai Bonet, Hector Geffner

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

Reproducibility Variable Result LLM Response
Research Type Experimental 7 Experimental Results The general PBT algorithm can use different algorithms for computing marginals from the factors. We experiment with the jointree (JT), belief propagation (BP), and ACm algorithms for m 2 {0, 1}. For JT and BP we use libdai [Mooij, 2010] while ACm is ours. The experiments were performed on Intel Xeon E5-2666 CPUs running at 2.9GHz with a memory cap of 10Gb (exhausted/approached only by JT).
Researcher Affiliation Academia Blai Bonet Universidad Sim on Bol ıvar Caracas, Venezuela bonet@ldc.usb.ve Hector Geffner ICREA & Universitat Pompeu Fabra Barcelona, SPAIN hector.geffner@upf.edu
Pseudocode No The paper describes algorithms (JT, BP, ACm, PBT) but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper mentions using 'libdai' which is described as 'a free and open source C++ library for discrete approximate inference in graphical models', but this is a third-party tool used by the authors, not the source code for their own PBT methodology.
Open Datasets No The paper describes problems like 'Minesweeper', '1-line-3-SLAM', and 'Minemapping' and their formulations. It refers to 'Murphy's 1-line SLAM problem [1999]' as a problem formulation in a referenced paper, not a publicly accessible dataset with explicit access information (link, DOI, repository, or citation with authors/year for a dataset itself).
Dataset Splits No The paper mentions 'averages over 500 runs' or 'averages over 100 random executions' as part of the experimental methodology, but it does not specify training, validation, and test dataset splits in terms of percentages, sample counts, or references to predefined splits for data partitioning.
Hardware Specification Yes The experiments were performed on Intel Xeon E5-2666 CPUs running at 2.9GHz with a memory cap of 10Gb (exhausted/approached only by JT).
Software Dependencies No The paper mentions using 'libdai [Mooij, 2010]' for JT and BP, but it does not specify the version number for libdai or any other software dependency relevant to reproducibility.
Experiment Setup Yes Table 2 shows results for instances of size n = 64 and n = 512, using 16 and 256 particles sampled with the optimal proposal distribution. The executions choose actions randomly until each cell of the grid is visited 10 times. For this problem, a cell is assumed labeled when for one of the possible colors its marginal probability is 0.55 of higher.