Reviews: Comparison Against Task Driven Artificial Neural Networks Reveals Functional Organization in Mouse Visual Cortex

Strengths: I found the authors' formulation of network pseudo-depth to be a very interesting and potentially useful metric for comparing artificial neural network models to neural data. I think their finding (displayed in Figure 2) that the number of sampled neurons had to be around at least 1000-2000 for the VGG-16 pseudo-depth to be consistently estimated, and that this finding holds when comparing representations against another network (VGG-19), demonstrates a potentially useful rule-of-thumb for adequate population sizes in neural data. Furthermore, their finding that mouse visual cortex is more parallel after a few stages of hierarchical processing starting at around area VISp, could be useful for building better task-driven models of mouse visual cortex, and indicates an important distinction with the traditional, hierarchical primate ventral visual pathway. Weaknesses: I would have liked to see more analyses of the robustness of the pseudo-depth metric with different networks, especially those not in the VGG family. I am aware that the Allen Institute has compared VGG-16/19 to their mouse data, and therefore, this is likely why the authors chose this model to begin with.