Learning is accelerated across various pre-training dataset variations.

<b>A.</b> Example datapoint and learning trajectories on the propagating square dataset (Methods <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1012830#sec012" target="_blank">4.1.3</a>). Learning trajectories of four vanilla RNN models on the simulated corridor training (left) and validation (right) datasets. Shown is the average MSE across five-fold cross-validation and the corresponding standard deviation. Training continues beyond the epochs shown, with each model stopping individually once overfitting is detected. <b>B.</b> Same as panel A, but for the pre-training dataset consisting of retinal waves with added Gaussian blur (Methods <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1012830#sec010" target="_blank">4.1.1</a>). <b>C.</b> Same as panel A, but for the pre-training dataset comprising retinal waves (RWs) that propagate in all directions with a directional bias (Methods <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1012830#sec010" target="_blank">4.1.1</a>). (PDF)