Understanding the computational mechanisms of human vision therefore requires us to measure and model the rapid representational dynamics across the different regions of the ventral stream. These give rise to recurrent interactions, which are thought to contribute to visual inference ( 3– 13). However, the primate visual system contains abundant lateral and feedback connections ( 2). In human neuroscience and corresponding modeling work, insight has often been generated based on time-averaged data and feedforward computational models. Although considerable progress has been made in characterizing the neural selectivity across much of the system, the underlying computations are not well understood. Vision relies on an intricate network of interconnected cortical regions along the ventral visual pathway ( 1). These results establish that recurrent models are required to understand information processing in the human ventral stream. Targeted virtual cooling experiments on the recurrent deep network models further substantiate the importance of their lateral and top-down connections. Finally, recurrent deep neural network models clearly outperform parameter-matched feedforward models in terms of their ability to capture the multiregion cortical dynamics. Categorical divisions emerge in sequence, cascading forward and in reverse across regions, and Granger causality analysis suggests bidirectional information flow between regions. We observe substantial representational transformations during the first 300 ms of processing within and across ventral-stream regions.
Here, we measure and model the rapid representational dynamics across multiple stages of the human ventral stream using time-resolved brain imaging and deep learning. Despite its abundant lateral and feedback connections, object processing is commonly viewed and studied as a feedforward process. The human visual system is an intricate network of brain regions that enables us to recognize the world around us.
0 kommentar(er)
