Discriminative Structure Learning of Arithmetic Circuits

Abstract

The biggest limitation of probabilistic graphical models is the complexity of inference, which is often intractable. An appealing alternative is to use tractable probabilistic models, such as arithmetic circuits (ACs) and sum-product networks (SPNs), in which marginal and conditional queries can be answered efficiently. In this paper, we present the first discriminative structure learning algorithm for ACs, DACLearn (Discriminative AC Learner), which optimizes conditional log-likelihood. Based on our experiments, DACLearn learns models that are more accurate and compact than other tractable generative and discriminative baselines.

Cite

Text

Rooshenas and Lowd. "Discriminative Structure Learning of Arithmetic Circuits." International Conference on Artificial Intelligence and Statistics, 2016. doi:10.1609/aaai.v30i1.9963

Markdown

[Rooshenas and Lowd. "Discriminative Structure Learning of Arithmetic Circuits." International Conference on Artificial Intelligence and Statistics, 2016.](https://mlanthology.org/aistats/2016/rooshenas2016aistats-discriminative/) doi:10.1609/aaai.v30i1.9963

BibTeX

@inproceedings{rooshenas2016aistats-discriminative,
  title     = {{Discriminative Structure Learning of Arithmetic Circuits}},
  author    = {Rooshenas, Amirmohammad and Lowd, Daniel},
  booktitle = {International Conference on Artificial Intelligence and Statistics},
  year      = {2016},
  pages     = {1506-1514},
  doi       = {10.1609/aaai.v30i1.9963},
  url       = {https://mlanthology.org/aistats/2016/rooshenas2016aistats-discriminative/}
}