Encoding-state-triggered stimulation

A closed-loop paradigm that continuously decodes a subject's momentary memory-encoding state from ongoing neural activity and delivers stimulation only when the brain is predicted to be encoding poorly. A classifier trained on the neural correlates of subsequently-remembered versus subsequently-forgotten items (a subsequent-memory biomarker) drives, in real time, brief stimulation of a memory-supporting region such as lateral temporal cortex. Under the DARPA Restoring Active Memory program this approach reportedly improved word recall when stimulation was gated to predicted low-memory states, whereas the same stimulation applied during good-encoding states could be neutral or harmful.

The design principle — stimulate the deficit, not the clock — generalizes well beyond memory and is a template for state-dependent neuromodulation. Caveats mirror the field: small epilepsy cohorts, laboratory word-list tasks, biomarkers that may not transfer across people or tasks, and effect sizes that are statistically real but modest relative to the goal of restoring clinically meaningful memory.