Expand description
Distributional refinement of atom vectors.
Atoms start as random vectors (deterministic from string hash). Over time, atoms that appear in similar relational contexts (same relations, same co-occurring peers) should become more similar. This module refines atom vectors by blending them toward their abstraction prototypes, producing self-organizing semantic embeddings without any external model.
The refinement step runs as part of the cognition loop or is called explicitly via the Governor. It requires write access to AtomMemory.
Structsยง
- Distributional
Refiner - Refines atom vectors by blending them toward distributional context vectors.
- Refinement
Report - Result of a refinement pass.
- Refiner
Config - Configuration for distributional refinement.