How well do deep neural networks fare as models of mouse visual cortex? A majority of research to date suggests results far more mixed than those produced in the modeling of primate visual cortex. Here, we perform a large-scale benchmarking of dozens of deep neural network models in mouse visual cortex with both representational similarity analysis and neural regression. Using the Allen Brain Observatory's 2-photon calcium-imaging dataset of activity in over 6,000 reliable rodent visual cortical neurons recorded in response to natural scenes, we replicate previous findings and resolve previous discrepancies, ultimately demonstrating that modern neural networks can in fact be used to explain activity in the mouse visual cortex to a more reasonable degree than previously suggested. Using our benchmark as an atlas, we offer preliminary answers to overarching questions about levels of analysis (e.g.

Moreover, only a finite number of stimuli can be shown to an animal over the course of a realistic experiment.

Figure 1b shows the average noise ceiling values for every area. As noted in section 3.3, the CPC loss function relies on predicting the future latent representations ANNs that are trained with two other supervised loss functions: supervised object categorization (ImageNet dataset) and supervised action recognition (UCF101 dataset). All the models are compared based on their representation alignment with different areas of mouse visual cortex (see section 3.2), and by using downstream tasks (see For training the deep ANNs, we use the UCF101 dataset. Brain Observatory presented to the mice. As noted in section 3.3, we examine the two pathways of our trained ResNet-2p on two downstream For each downstream task, the weights of the trained pathways are frozen and a linear classifier is trained on the final convolutional layer of each pathway.

In particular, the clearest example is the specialization of the ventral ("what") and dorsal

Deep predictive models of neuronal activity have recently enabled several new discoveries about the selectivity and invariance of neurons in the visual cortex.

How well do deep neural networks fare as models of mouse visual cortex?

However, existing DNNs are mostly designed to analyze neural responses to static images, relying on feedforward structures and lacking physiological neuronal mechanisms. There is limited insight into how the visual cortex represents natural movie stimuli that contain context-rich information.

Neural Information Processing Systems http://nips.cc/