Representational manifolds in olfactory cortex

Piriform cortex (PCx) and its zebrafish homolog, telencephalic area pDp, have been proposed to function as autoassociative networks that learn internal representations of relevant odor categories. Recent results indicate that odor-evoked activity is organized on neural manifolds that differ from discrete attractor states in classical models. These manifolds are disentangled by experience and maintain generalizable representations by restructuring representational space itself. Modeling suggests that representational geometry is shaped, at least partially, by recurrent connectivity in inhibition-stabilized networks. Here, subtle deviations from perfect excitation-inhibition balance warp representational manifolds without compromising stability. Representational drift on multiple timescales, along with the emergence of joint olfactory and non-olfactory representations following behavioral training, further suggest that PCx/pDp encode not only odor identity but also temporal context and multimodal associations along distinct manifold dimensions. These findings support a view in which PCx/pDp reorganize representations through changes in manifold geometry, thereby supporting distributed and continual learning.