Integrating Deep Learning Based Perception with Probabilistic Logic via Frequent Pattern Mining

Authors: Ben Goertzel, Ted Sanders, Jade O'Neill, Gino Yu
Year: 2013
Venue: Artificial General Intelligence — AGI 2013. Lecture Notes in Computer Science, vol 7999. Springer. pp. 40-49.
DOI: 10.1007/978-3-642-39521-5_5

Summary

Bridges deep learning perception (DeSTIN belief-tree representation) with PLN reasoning through frequent pattern mining over Atomspace-translated perceptual states. The mining stage is described as Frequent Subtree Mining over Yun Chi's external software (Chi et al., IEEE TKDE 2005, ref [13]).

Relevance to Hyperon

Demonstrates the neural-symbolic integration approach that recurs in fragmented form across Hyperon-era research — connecting deep learning perception with logical reasoning via pattern mining. The 2013 architecture (DeSTIN → frequent-pattern miner → PLN) is now [PARTIAL-FRAGMENTED-REVIVAL]: layer fragments exist (perception via opencog/sensory; mining via the hyperon-miner trio + neural-subgraph-matcher-miner; inference via hyperon-pln + chaining) but the layers are not interconnected at runtime. See Perception/Neural-Symbolic cluster pilot 2026-05-01.

Editorial note on terminology

The 2013 paper does not use the term "FISHGRAM" anywhere in its title or body. "FISHGRAM" is an OpenCog-Python implementation name (sentence-case "Fishgram") at opencog/python/learning/fishgram/fishgram.py in opencog-singnet. The paper itself cites Yun Chi's Frequent Subtree Mining software as the algorithmic source. Wiki cards and synthesis documents that use "FISHGRAM" as shorthand for the 2013 paper architecture are using the term editorially, not citing the paper's own vocabulary.

Provenance

Source PDF retrieved 2026-04-30 from author-uploaded conference version (goertzel.org/agi-13/DeSTIN_PLN_v3.pdf);cross-referencedpublication_texts/2013_Goertzel_Sanders_ONeill_Yu_DeSTIN_PLN.pdf against Springer DOI and ResearchGate (publication 262157725). Local copy: .