Compositional Models: Iterative Structure Learning from Data

It is well known that Bayesian network structure learning from data is an NP-hard problem. Learning a correct skeleton of a DAG is the foundation of dependency analysis algorithms for this problem. Considering the unreliability of the high order condition independence (CI) tests and the aim to improve the efficiency of a dependency analysis algorithm, the key steps are to use less number of CI tests and reduce the sizes of condition sets as many as possible. Based on these analyses and inspired by the algorithm HPC, we present an algorithm, named efficient hybrid parents and child (EHPC), for learning the adjacent neighbors of every variable. We proof the validity of the algorithm. Compared with state-of-the-art algorithms, the experimental results show that EHPC can handle large network and has better accuracy with fewer number of condition independence tests and smaller size of conditioning set.

Download Full-text

A survey on Bayesian network structure learning from data

Progress in Artificial Intelligence ◽

10.1007/s13748-019-00194-y ◽

2019 ◽

Vol 8 (4) ◽

pp. 425-439 ◽

Cited By ~ 7

Author(s):

Mauro Scanagatta ◽

Antonio Salmerón ◽

Fabio Stella

Keyword(s):

Bayesian Network ◽

Network Structure ◽

Structure Learning ◽

Learning From Data ◽

Bayesian Network Structure ◽

Bayesian Network Structure Learning

Download Full-text

Online Identification of Multivariable Nonlinear Processes with Iterative Structure Learning and Application to a Diesel Engine

IFAC Proceedings Volumes ◽

10.3182/20140824-6-za-1003.02641 ◽

2014 ◽

Vol 47 (3) ◽

pp. 9216-9222 ◽

Cited By ~ 1

Author(s):

Marek Kowalczyk ◽

Boliang Yi ◽

Rolf Isermann

Keyword(s):

Diesel Engine ◽

Structure Learning ◽

Nonlinear Processes ◽

Online Identification ◽

Iterative Structure

Download Full-text

A Structure Learning Method for Block-Based Neural Networks

IEEJ Transactions on Electronics Information and Systems ◽

10.1541/ieejeiss.133.1976 ◽

2013 ◽

Vol 133 (10) ◽

pp. 1976-1982 ◽

Cited By ~ 3

Author(s):

Hidetaka Watanabe ◽

Seiichi Koakutsu ◽

Takashi Okamoto ◽

Hironori Hirata

Keyword(s):

Neural Networks ◽

Structure Learning ◽

Learning Method ◽

Block Based

Download Full-text

Pushing the Envelope

10.1093/oso/9780199670871.003.0014 ◽

2018 ◽

Author(s):

Eaton E. Lattman ◽

Thomas D. Grant ◽

Edward H. Snell

Keyword(s):

High Resolution ◽

Electron Density ◽

Structural Information ◽

Hydration Shell ◽

Three Dimensional ◽

Scattering Data ◽

Saxs Data ◽

Electron Density Function ◽

Angle Scattering ◽

Iterative Structure

Direct electron density determination from SAXS data opens up new opportunities. The ability to model density at high resolution and the implicit direct estimation of solvent terms such as the hydration shell may enable high-resolution wide angle scattering data to be used to calculate density when combined with additional structural information. Other diffraction methods that do not measure three-dimensional intensities, such as fiber diffraction, may also be able to take advantage of iterative structure factor retrieval. While the ability to reconstruct electron density ab initio is a major breakthrough in the field of solution scattering, the potential of the technique has yet to be fully uncovered. Additional structural information from techniques such as crystallography, NMR, and electron microscopy and density modification procedures can now be integrated to perform advanced modeling of the electron density function at high resolution, pushing the boundaries of solution scattering further than ever before.

Download Full-text