Learning the Structure of Sum-Product Networks via an SVD-based Algorithm
- Submitting institution
-
University of Glasgow
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 11-09962
- Type
- E - Conference contribution
- DOI
-
-
- Title of conference / published proceedings
- 31st Conference on Uncertainty in Artificial Intelligence (UAI 2015)
- First page
- 32
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- July
- Year of publication
- 2015
- URL
-
http://eprints.gla.ac.uk/215100/
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
2
- Research group(s)
-
-
- Citation count
- 10
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- ORIGINALITY: Presents one of the first algorithms on learning the structure of sum-product networks (graphical models with inference tractability advantages) using rank-1 submatrix extraction. SIGNIFICANCE: Represents one of the leading algorithms in the structure learning of sum-product networks (SPNs), producing better likelihood and inference results and much greater speed based on rank-1 submatrix extraction, compared to previous local splitting approaches. Uses correlations which are easier to estimate than independences. Presented at a top AI conference. RIGOUR: The algorithm was developed on rigorous mathematical foundations, and empirically tested on over 20 datasets. Theoretical guarantees are provided for algorithm performance scaling.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -