Flexible decision-making in recurrent neural networks trained with a biologically plausible rule

Neural activity during cognitive tasks exhibits complex dynamics that flexibly encode task-relevant variables. Recurrent neural networks operating in the near-chaotic regime, which spontaneously generate rich dynamics, have been proposed as a model of cortical computation during cognitive tasks. However, existing methods for training these networks are either biologically implausible, and/or require a continuous, real-time error signal to guide the learning process. The lack of a biological learning method currently restricts the plausibility of recurrent networks as models of cortical computation. Here we show that a biologically plausible learning rule can train such recurrent networks, guided solely by delayed, phasic rewards at the end of each trial, for nontrivial tasks.