How does the brain maintain stable internal representations of the world while also remaining flexible enough to learn, adapt, and predict the future? This is a central question in systems neuroscience. In this talk, I will present recent work from my lab that addresses this problem at both the level of network dynamics and the level of single-neuron coding.
First I will focus on the head direction system and the concept of network gain as a control parameter that regulates how strongly external landmarks can realign an internal directional attractor during reorientation. I will describe new data and models that begin to reveal the circuit mechanisms that may tune this gain signal in different behavioral contexts.
I will then turn to hippocampal representations across days. Using chronic recordings, we find that not all place cells drift. In one set of experiments, a majority of cells whose firing is constrained by environmental geometry can remain remarkably stable across sessions, whereas other cells show substantial drift. In a separate set of experiments, we identify a distinct population of reward coding neurons whose drift is not random but highly organized, showing a systematic backward shift over learning as an activity transitions from encoding current outcomes to predicting future reward locations. Together, these studies begin to specify when neural representations are stable, when they change, and how that change can follow precise trajectories rather than random wander, providing concrete constraints for models of long-term spatial memory and its disruption in disease.