This paper proposes a new classification model called logistic circuits. On MNIST and Fashion datasets, our learning algorithm outperforms neural networks that have an order of magnitude more parameters. Yet, logistic circuits have a distinct origin in symbolic AI, forming a discriminative counterpart to probabilistic-logical circuits such as ACs, SPNs, and PSDDs. We show that parameter learning for logistic circuits is convex optimization, and that a simple local search algorithm can induce strong model structures from data.

@inproceedings{LiangAAAI19,
  author    = {Liang, Yitao and Van den Broeck, Guy},
  title     = {Learning Logistic Circuits},
  booktitle = {Proceedings of the 33rd Conference on Artificial Intelligence (AAAI)},
  month     = 1,
  year      = {2019},
  url       = "http://starai.cs.ucla.edu/papers/LiangAAAI19.pdf",
  slides    = "http://starai.cs.ucla.edu/slides/AAAI19-logistic-circuits.pdf",
  code = "https://github.com/UCLA-StarAI/LogisticCircuit",
  annotation = "(Oral full presentation, acceptance rate 460/7700 = 6\%)",
  keywords  = {conference,selective}
}