research · kei-neuromem

kei-neuromem — a memory that grows.

A growing associative memory for language that stores only the connections it is confident about, and feeds a conventional downstream model instead of trying to be the model.

The idea

Instead of one large model trying to represent language at every scale at once, kei-neuromem builds a growing memory of the connections it is confident about and hands them to a conventional model downstream. The memory is a module that feeds a downstream processor, not a replacement for one.

Built not to collapse

The memory is designed on a stability principle that lets it hold many distinct items at once, within its capacity, instead of collapsing them into a single one — a property of the design rather than of tuning.