Abstract

Neural representations change over time, even for habitual behaviors. This phenomena, termed “representational drift”, seems to be at odds with long-term stable neural representations. Previously, we showed that representational drift was gradual, and might be tracked using weak error feedback. In this talk, I show how stable representations could be achieved without external error feedback. I’ll discuss a model for representational drift, which captures features of neural population codes observed experimentally: Tunings are typically stable, but occasionally undergo larger reconfigurations. I then discuss “self healing codes”, which combine error-correction with neural plasticity. Self-healing codes can track drift without outside error feedback. The learning rule required is biologically plausible, and amounts to a form of homeostatic Hebbian plasticity. When combined with network interactions that allow neurons to share information, such homeostatic plasticity could allow a subpopulation of stable cells to maintain an accurate readout of an unstable population code.