Sum-Product Networks

The Sum-Product Network (SPN) is a new type of machine learning model with fast exact probabilistic inference over many layers.

Advantages of SPNs:

  • Unlike graphical models, SPNs are tractable over high treewidth models
  • SPNs are a deep architecture with full probabilistic semantics
  • SPNs can incorporate features into an expressive model without requiring approximate inference.

SPNs have achieved impressive results on numerous datasets, including:

  • Image completion
  • Image classification
  • Activity recognition
  • Click-through logs
  • Nucleic acid sequences
  • Collaborative filtering