Abstract
Perception depends on a complex interplay between feedforward and recurrent processing. Yet, while the former has been extensively characterized, the computational organization of the latter remains largely unknown. Here, we use magneto-encephalography to localize, track and decode the feedforward and recurrent processes of reading, as elicited by letters and digits whose level of ambiguity was parametrically manipulated. We first confirm that a feedforward response propagates through the ventral and dorsal pathways within the first 200 ms. The subsequent activity is distributed across temporal, parietal and prefrontal cortices, which sequentially generate five levels of representations culminating in action-specific motor signals. Our decoding analyses reveal that both the content and the timing of these brain responses are best explained by a hierarchy of recurrent neural assemblies, which both maintain and broadcast increasingly rich representations. Together, these results show how recurrent processes generate, over extended time periods, a cascade of decisions that ultimately accounts for subjects' perceptual reports and reaction times.
Highlights
To process the rich sensory flow emanating from the retina, the brain recruits a hierarchical network originating in the primary visual areas and culminating in the infero-temporal, dorso-parietal and prefrontal cortices (Hubel and Wiesel, 1962; Maunsell and van Essen, 1983; Riesenhuber and Poggio, 1999; DiCarlo et al, 2012).In theory, the feedforward recruitment of this neural hierarchy could suffice to explain our ability to recognize visual objects
We show that the late and sustained neural activity distributed along the visual pathways generates, over extended time periods, a cascade of categorical decisions that predicts subjects’ perceptual reports
We show with source-localized MEG that recurrent processes sequentially generate, over an extended time period, a hierarchy of representations that account for the timing and the content of perceptual reports
Summary
The feedforward recruitment of this neural hierarchy could suffice to explain our ability to recognize visual objects. Recent studies demonstrate that artificial feedforward neural networks trained to categorize objects generate similar activations patterns to those elicited in the infero-temporal cortices (Yamins et al, 2014; Schrimpf et al, 2019; Khaligh-Razavi and Kriegeskorte, 2014; Cichy et al, 2016). Feedforward architectures have a fixed number of processing stages and are unable to explain a number of neural and perceptual phenomena. The neural responses to visual stimuli generally exceed the 200 ms feedforward recruitment of the visual hierarchy (Dehaene and Changeux, 2011a; Lamme and Roelfsema, 2000)
Sign-in/Register to view highlights & summary 
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call