Matthijs Pals

The paper uses data-constrained RNNs to analyze neural activity in macaques and finds attractor dynamics underlie sequence working memory.

The paper introduces JO_SCPLOWAXLEYC_SCPLOW, a framework for differentiable simulation of biophysical neuron models, enabling efficient parameter optimization and task-driven training of large-scale models.

The paper examines whether the phase of neuronal firing reflects temporal order in sequence memory in humans and RNNs, finding that phase order does not match item order.

This paper shows that trained recurrent neural networks develop phase-locked limit cycles to support working memory through coupling with an external reference oscillation.