Skip to content
mcj-groupPublic

About

Parallel priority-ordered belief propagation code used in [PGM 2026] and [SPAA 2024]

Topics

Resources

Stars

3 stars

Watchers

0 watching

Forks

 
 

Latest commit

 

History

111 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Relaxed Residual Belief Propagation with the Stealing Multiqueue and Lazy Priority Updates

@inproceedings{bassi:pgm26rbpsmq,
    title     ={Optimized Priority Scheduling for Faster Scalable Belief Propagation},
    author    ={Bassi, Abnash and Posluns, Gilead and Jeffrey, Mark C.},
    booktitle ={Proc. 13th International Conference on Probabilistic Graphical Models},
    series    ={PGM},
    year      ={2026},
    url       ={https://openreview.net/forum?id=iATKlKvddx}
}

Steps to reproduce experiment results

Setup and Compilation

  1. On a Debian/Ubuntu machine, install the Boost library using sudo apt install libboost-all-dev. If you are using a different OS, follow the corresponding instructions here. We recommend Boost version 1.85.0.
  2. mkdir golden to store residual algorithm results for accuracy comparisons
  3. rm -rf build
  4. make

Evaluation

Examples:

./build/main residual ising default 2
./build/main relaxed-rbp-mq ising default 48 2
./build/main relaxed-rbp-smq ising default 48 2
./build/main synch_bp_mt ising default 48 2

General command syntax:

./build/main <algorithm> <mrf> <size> [all except residual: <threads>] [optional: <seed>]

Argument Options:

algorithms: residual, relaxed-rbp-mq, relaxed-rbp-smq, relaxed-smart-splash-mq, relaxed-smart-splash-smq, synch_bp_mt

mrf: ising, potts, ldpc, tree

size: default to match paper results or other value

threads: machine-dependent

seed: 2 to match paper results or other value

About

Parallel priority-ordered belief propagation code used in [PGM 2026] and [SPAA 2024]

Topics

Resources

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages