Visual working memory representations bypass divisive normalization

Abstract

How distinct is visual working memory from visual perception? Although evidence suggests that briefly remembered stimuli are represented within early visual cortices, it remains unknown whether these visual memory traces undergo the same neural computations as true visual representations. Here, we examined whether visual memory representations succumb to a hallmark visual computation: divisive normalization. The strength of normalization was assessed during perception by utilizing a surround suppression paradigm wherein participants replicated the contrast of a center stimulus, which was enveloped by a full contrast surround. We measured perceived contrast for five contrast levels (10-75% contrast) by means of a method-of-adjustment task. Observers were presented with a center-surround stimulus and were asked after a 2s retention interval to use a dial to match the contrast held in memory. Results revealed substantial suppression of the apparent contrast across all contrast levels. Next, we moved the surround stimulus into the retention interval to test the degree to which a contrast memory representation can be influenced by a subsequently presented surrounding stimulus. If representations within visual working memory undergo normalization similarly to visual representations, the presence of the surrounding stimulus should attenuate the apparent contrast of the remembered stimulus, just as it would when center and surround stimuli are presented simultaneously. We quantified this by fitting perceived contrast estimates with the normalization model, and found evidence to the contrary: while the presence of the surround stimulus attenuated memory for the contrast of the remembered stimulus when they were presented simultaneously, the presence of the surround stimulus during the retention interval had no impact on the quality of visual memory, indicating a lack of contrast normalization between working memory stores. While visual memory representations may reside within early visual areas, they are qualitatively distinct from true visual representations, bypassing divisive normalization, a canonical visual computation.

Meeting abstract presented at VSS 2017