,
15 tweets,
12 min read
Read on Twitter
Our new preprint on the ‘Neural dynamics of perceptual inference and its reversal during imagery’ with @LucaAmb and @marcelge is out now! We reveal feedforward and feedback dynamics of stimulus processing during perception and imagery. A thread 1/N biorxiv.org/content/10.110…
@LucaAmb @marcelge We then tested when these feedforward classifiers became active during imagery. Inspired by work from @MarWimber lab, we inferred the timing of perceptual classifier activation during imagery per trial using classifier distances. 4/N 
@LucaAmb @marcelge @MarWimber If stimulus processing happens in the same order, we expect a positive relation between perception classifier time and imagery reactivation time. If processing is reversed, we expect a negative relation, with late perception models being reactivated earlier during imagery. 5/N
@LucaAmb @marcelge @MarWimber We found a clear negative relationship between perception model time and imagery reactivation time (A). This indicates that the perceptual feedforward cascade is reactivated in reverse order during imagery with high-level areas becoming active before low-level areas (B). 6/N
@LucaAmb @marcelge @MarWimber Until this point we only focused on the initial feedforward sweep during perception. However, feedback processes are also assumed to play an important role in later stages of perception. Next, we looked at imagery reactivation of all time points during stimulus presentation. 7/N
@LucaAmb @marcelge @MarWimber The results hit us right between the eyes: during the first 400 ms of perception, there was a clear oscillatory pattern in the relationship between perception model time and imagery reactivation time, peaking at ~11 Hz. 8/N
@LucaAmb @marcelge @MarWimber Since the first 150 ms during perception reflects feedforward processing, which shows a negative relation with imagery, this pattern of alternating positive and negative slopes reflects recurrent feedforward (blue lines) and feedback (pink lines) processing during perception. 9/N
@LucaAmb @marcelge @MarWimber We then went on to validate this slow-wave recurrency within perception by determining the reactivation during perception of the identified feedforward and feedback phases. Here, negative slopes indicate reversals of information flow between different phases of perception. 10/N
@LucaAmb @marcelge @MarWimber We again found an oscillatory pattern of recurrence. However, interestingly, reversals of information flow were restricted to neighbouring phases, suggesting that stimulus representations changed over subsequent recurrent cycles (A-B). 11/N
@LucaAmb @marcelge @MarWimber Finally, we checked whether these feedforward and feedback phases could be inferred based on evoked oscillations and found a peak at 10 Hz (C). This further emphasizes that recurrent processing during perception is aligned to the alpha frequency. 12/N
@LucaAmb @marcelge @MarWimber This also means that this new analysis technique can be used to characterize recurrent processing during perception in other data sets without imagery to identify the different phases. Get in touch if you’re interested! 13/N
@LucaAmb @marcelge @MarWimber In conclusion, we found that the initial perceptual feedforward sweep is reversed during imagery and that later stages of perception are characterized by recurrent processing up and down the visual hierarchy aligned to an alpha frequency. 14/N
@LucaAmb @marcelge @MarWimber These results are in line with the idea that during perception, high-level causes of sensory input are inferred through recurrent hypothesis testing whereas during imagery, this inferred mapping is reversed to generate sensory representations given high-level concepts. N/N
