Neuromorphic computing · Relationship discovery
A neural system for exploring Scripture-scale relationships.
EION Labs developed a custom spiking neural research system to investigate latent relationships across a shared memory space representing all 31,102 verses of the King James Version.
Research question
What relationships might a different kind of neural system reveal?
Keyword search finds exact language. Curated cross-references preserve known connections. Embedding similarity identifies related representations. EION’s neuromorphic research asks what additional patterns may emerge when Scripture memories interact through an event-driven neural structure.
The objective is exploratory: surface candidate associations that may deserve examination, then return them to the biblical text and Arca’s deterministic evidence architecture for verification.
Research system
The exported artifact records a complete verse-memory space and the sparse neural structure learned during training. It is a research instrument, not a theological authority or autonomous interpreter.
Shared verse space
Every KJV verse is represented by a 64-dimensional memory, creating a common input space for the complete 31,102-verse canon.
Leaky integrate-and-fire
128 simulated neurons integrate incoming signals over discrete time, decay prior state, emit spikes at a threshold, and reset.
Pairwise STDP
Spike-timing-dependent plasticity adjusts synaptic relationships according to the relative timing of neural activity.
Learned connectivity
After 33 training epochs, the artifact retains sparse recurrent and input connectivity for relationship investigation and ranking research.
Discovery and verification
The neural layer and Arca perform different jobs. Their separation is a research safeguard, not an implementation detail.
Represent
A verse enters the shared memory space.
Activate
Event-driven activity propagates through the learned structure.
Surface
Candidate relationships are ranked for investigation.
Verify
Arca returns each candidate to canonical text and typed evidence.
A similarity is not an interpretation, and a spike is not authority. The graph provides verification; the neural system helps explore where verification may be fruitful.
Current status
An active experimental research layer.
The system demonstrates a complete trained memory artifact and a production deterministic ranker for investigating learned associations. Formal comparison across neural, lexical, graph, and hybrid retrieval configurations remains ongoing.
Precise transformations, training procedures, thresholds, ranking combinations, and production strategies remain proprietary. Public documentation describes the research architecture and its limits without publishing a reproduction recipe.
Neuromorphic output identifies candidates for examination. It does not establish doctrine, determine meaning, or replace reading Scripture in context.