Residual Splash for Optimally Parallelizing Belief Propagation

By Joseph E. Gonzalez, Yucheng Low, Carlos Guestrin. 2009

141 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Kernel Belief Propagation (2011, methodology): “We used the Splash scheduling of Gonzalez et al. (2009) since it provided the lowest runtime among all tested schedulings.”
Distributed GraphLab: A Framework for Machine Learning in the Cloud (2012, methodology): “, belief propagation [13], expectation maximization [28], and stochastic optimization [35, 34]) where asynchronous procedures have been empirically shown to significantly outperform synchronous procedures.”
Distributed GraphLab: A Framework for Machine Learning in the Cloud (2012, methodology): “ For example, linear systems (common to many MLDM algorithms) have been shown to converge faster when solved asynchronously [4]. Additionally, there are numerous other cases (e.g., belief propagation [13], expectation maximization [28], and stochastic optimization [35, 34]) where asynchronous proced”
GraphLab: A New Framework For Parallel Machine Learning (2010, methodology): “The Systolic abstraction also cannot express the dynamic and specialized structured schedules which were shown by Gonzalez et al. [2009a,b] to dramatically improve the performance of algorithms like BP.”
Recursive regularization for large-scale classification with hierarchical and graphical dependencies (2013, background): “For related issues, refer [14] which discuss parallelizing belief propagation on graphs.”
GraphLab: A Distributed Framework for Machine Learning in the Cloud (2011, background): “Unfortunately, transforming ML algorithms with computational dependencies into the embarrassingly parallel form needed for these abstractions is often complicated and can introduce substantial algorithmic inefﬁciency [19].”
Stochastic variational inference for hidden Markov models (2014, background): “Buffering subchains is related to splash belief propagation (BP) for parallel inference in undirected graphical models, where the belief at any given node is monitored based on locally-aware message passing in order to maintain a good approximation to the true belief [21].”
An architecture for parallel topic models (2010, unknown): “This work is complementary to recent progress on efficient inference in graphical models [5].”
PowerGraph: Distributed Graph-Parallel Computation on Natural Graphs (2012, unknown): “In [18, 19, 29] we demonstrated that asynchronous computation can lead to both theoretical and empirical”
Distributed message passing for large scale graphical models (2011, methodology): “In our experiments, the roundrobin scheduler of GraphLAB did not perform as well as SplashBP.”
Polonium: Tera-Scale Graph Mining and Inference for Malware Detection (2011, background): “al [30] have developed a parallelized version of Belief Propagation that runs on a multi-core, sharedmemory framework, which unfortunately precludes us from readily applying it on our problem, as our current graph does not fit in memory.”