Elastic Associative Memory (EAM)

The memory model HeatherDB implements: an associative store whose structure reshapes itself as data arrives.

Like you're twelve

Imagine a shelf that rearranges itself while you use it. Put three cookbooks on it and it quietly slides them together; add a pile of maps and it makes a region for those too, moving the cookbooks a little to make room. Nobody reorganises the shelf on a Sunday. The rearranging is how things get put away.

That elasticity is the difference between a memory that learns and a filing cabinet that fills.

Precisely

EAM is an Associative memory whose representational capacity and internal organisation adapt to the data written into it, rather than being fixed at build time. Writes are online learning steps: each one places a pattern, adjusts the neighbourhood it lands in, and updates the navigable structure used to reach it.

The consequences are the ones the paradigm advertises — no separate index to rebuild, no retraining to absorb drift, no ceiling declared before the data arrives, and no Catastrophic forgetting when new patterns differ from old.

The mechanism is described in How the memory works, and the formal account is in the paper linked from the documentation index.

Related

Associative memory · Online learning · Sparse Distributed Memory (SDM) · Pentti Kanerva