Considerations in using recurrent neural networks to probe neural dynamics
Recurrent neural networks (RNNs) are increasingly being used to model complex cognitive and motor tasks performed by behaving animals. Here, RNNs are trained to reproduce animal behavior while also recapitulating key statistics of empirically recorded neural activity. In this manner, the RNN can be viewed as an in silico circuit whose computational elements share similar motifs with the cortical area it is modeling. Further, as the RNN's governing equations and parameters are fully known, they can be analyzed to propose hypotheses for how neural populations compute. In this context, we present important considerations when using RNNs to model motor behavior in a delayed reach task. First, by varying the network's nonlinear activation and rate regularization, we show that RNNs reproducing single neuron firing rate motifs may not adequately capture important population motifs. Second, by visualizing the RNN's dynamics in low-dimensional projections, we demonstrate that even when RNNs recapitulate key neurophysiological features on both the single neuron and population levels, it can do so through distinctly different dynamical mechanisms. To militate between these mechanisms, we show that an RNN consistent with a previously proposed dynamical mechanism is more robust to noise. Finally, we show that these dynamics are sufficient for the RNN to generalize to a target switch task it was not trained on. Together, these results emphasize important considerations when using RNN models to probe neural dynamics.