Conductance-based synapses + spike correlations fix brain model paradox

A team led by Vicky Zhu, Gabriel Ocker, and Robert Rosenbaum has published a preprint on arXiv demonstrating empirical scaling laws in balanced recurrent networks with conductance-based synapses. Their simulations address a long-standing tension in computational neuroscience: current-based synapse models produce realistic variability only when paired with spike-time correlations, but conductance-based models alone yield unrealistically low membrane potential volatility. By introducing both realistic conductance dynamics and correlated spike timing simultaneously, the authors show that the two effects cancel out, generating moderate—and biologically plausible—levels of variability.

This result builds on recent work in feedforward networks and extends it to recurrent circuits, the dominant architecture in cortex. The paper suggests that modelers must include multiple realistic assumptions together to capture neural dynamics accurately. For AI researchers, this hints at why simple approximations (like current-based synapses) often fail to replicate cortical computations—and points toward more faithful neuromorphic designs. The arXiv preprint (2605.12404) is under review for publication.