Maanasa Natrajan

This paper shows that representational drift uncovers noise-robust neural representations that are difficult to learn but beneficial for memory stability.

This study uses place field repetition in rats to show that positional and nonpositional inputs to hippocampal CA1 are functionally dissociated, with positional inputs linked to spatial location and nonpositional inputs linked to temporal dynamics.

The paper shows that learning in the hippocampus produces orthogonalized neural representations resembling a state machine, captured by HMMs and biologically plausible networks, but not by gradient-based models.