Distributed Bayesian matrix factorization with limited communication

Xiangju Qin*, Paul Blomstedt, Eemeli Leppäaho, Pekka Parviainen, Samuel Kaski

*Tämän työn vastaava kirjoittaja

Tutkimustuotos: LehtiartikkeliArticleScientificvertaisarvioitu

3 Sitaatiot (Scopus)
148 Lataukset (Pure)

Abstrakti

Bayesian matrix factorization (BMF) is a powerful tool for producing low-rank representations of matrices and for predicting missing values and providing confidence intervals. Scaling up the posterior inference for massive-scale matrices is challenging and requires distributing both data and computation over many workers, making communication the main computational bottleneck. Embarrassingly parallel inference would remove the communication needed, by using completely independent computations on different data subsets, but it suffers from the inherent unidentifiability of BMF solutions. We introduce a hierarchical decomposition of the joint posterior distribution, which couples the subset inferences, allowing for embarrassingly parallel computations in a sequence of at most three stages. Using an efficient approximate implementation, we show improvements empirically on both real and simulated data. Our distributed approach is able to achieve a speed-up of almost an order of magnitude over the full posterior, with a negligible effect on predictive accuracy. Our method outperforms state-of-the-art embarrassingly parallel MCMC methods in accuracy, and achieves results competitive to other available distributed and parallel implementations of BMF.

AlkuperäiskieliEnglanti
Sivut1805–1830
JulkaisuMachine Learning
Vuosikerta108
Numero10
DOI - pysyväislinkit
TilaJulkaistu - 1 lokak. 2019
OKM-julkaisutyyppiA1 Julkaistu artikkeli, soviteltu

Sormenjälki

Sukella tutkimusaiheisiin 'Distributed Bayesian matrix factorization with limited communication'. Ne muodostavat yhdessä ainutlaatuisen sormenjäljen.

Siteeraa tätä