Abstract
Neuroplasticity enables the brain to adapt neural activity, but whether this can be harnessed for abstract optimization tasks like seeking curve extrema remains unclear. Here, we used a brain-machine interface in mice, pairing auditory feedback of neuronal firing rate with water rewards, to investigate whether motor cortex neurons can optimize activity along a unimodal curve ([Formula: see text]). The curve maps firing rate ([Formula: see text]) to sound frequency increase speed ([Formula: see text]), where the curve extremum accelerates reward acquisition. Over conditioning sessions, mice learned to modulate firing rates towards this peak, reducing reward time from 18.64 ± 7.30 s to 11.59 ± 4.38 s and increasing high-response events from 66 to 104 occurrences. Putative neurons increasingly prioritized high-response intervals, with positive proportion increments in upper intervals versus negative trends in lower ones. These findings demonstrate that cortical neurons can dynamically optimize activity along non-monotonic reward landscapes, revealing neuroplasticity as a substrate for adaptive self-optimization. This expands our understanding of how the brain learns abstract rules via feedback, with implications for neuroprosthetic design that leverage neural adaptability.