Fig. 3: Comparison of cross-validated prediction accuracy for encoding models based on task- and brain-optimized deep neural networks. | Nature Communications

Fig. 3: Comparison of cross-validated prediction accuracy for encoding models based on task- and brain-optimized deep neural networks.

From: Brain-optimized deep neural network models of human visual areas learn non-hierarchical representations

Fig. 3: Comparison of cross-validated prediction accuracy for encoding models based on task- and brain-optimized deep neural networks.The alternative text for this image may have been generated using AI.

a, b Accuracy / advantage plots for all subjects and brain areas for natural validation stimuli (a) and artificial stimuli (b). Format as in Fig. 1. c Average difference in model prediction accuracy (x-axis) for different levels of average prediction accuracy (y-axis) and for natural (thin curve) and artificial (thick curve) stimuli. d Average percentage of voxels for which the GNet-based encoding model explains more signal variance than AlexNet-based encoding model ("win percentage''; x-axis) for different levels of average prediction accuracy (y-axis). e Signal variance (%; y axes) in brain activity explained by varieties of network-based encoding models (colored bars; gray bars indicate theoretical upper limit) for all subjects (x-axis) in visual areas V1–V4. Inset: Average (across subjects) percent of explainable variance explained by a subset of the models. See Supplementary Table 1 for descriptions of all model acronyms. Source data are provided as a Source Data file.

Back to article page