Differentiable Probabilistic Logic Networks
Draft — This content has not been approved for publication.
← Back to Publications
Differentiable Probabilistic Logic Networks
Authors:
Year: 2019
Venue: Artificial General Intelligence (AGI 2019); arXiv:1907.04592
Links: arXiv abstract
Summary
Introduces a differentiable version of PLN whose rules operate over tensor truth values, so that a chain of reasoning steps builds a computation graph over tensors mapping premise truth values to conclusion truth values. This allows learning both premise truth values and rule formulas (expressed with trainable weights) by backpropagation, combining subsymbolic optimization with symbolic reasoning in the OpenCog cognitive-synergy spirit.
Relevance to Hyperon
A direct PLN extension that makes the framework differentiable and neural-compatible — relevant to the PLN Deep Dive's account of uncertain reasoning and to neural-symbolic integration / cognitive synergy.