Error-correcting dynamics in visual working memory

Abstract

Working memory is critical to cognition, decoupling behavior from the immediate world. Yet, it is imperfect; internal noise introduces errors into memory representations (1, 2). Such errors accumulate over time (3)(4)(5) and increase with the number of items simultaneously held in working memory (6-10). Here, we show that error-correcting attractor dynamics mitigate the impact of noise on working memory. These dynamics pull memories towards a few stable representations in mnemonic space, inducing a bias in memory representations but reducing the effect of noise. Model-based and model-free analyses show attractor dynamics account for the frequency, bias, and precision of working memory reports in both humans and monkeys. Furthermore, attractor dynamics were optimized to the context; they adapted to the statistics of the environment, such that memories drifted towards contextually-predicted values. Our results suggest attractor dynamics mediate errors in working memory by counteracting noise and integrating contextual information into memories.

Guidance of attention by working memory is a matter of representational fidelity

“…; or simply as a natural result of the fact that noise in memory is relatively independent for each item (Fougnie et al, 2012;Schurgin et al 2020;Panichello, DePasquale, Pillow, & Buschman, 2018;Wilken & Ma, 2004). Thus, variation between multiple working memory items appears to be an unavoidable and natural byproduct of working memory maintenance and so it is important to consider what this means for memory driven attentional guidance.…”

Build-up of serial dependence in color working memory

“…The build-up of serial dependencies during an experimental session has further implications for how we interpret their functional role. If serial dependence was an adaptation to exploit the world's tendency to remain stable 23 , and this adaptation could occur in the time scale of hour fractions (as recently shown for systematic biases in delayed-estimation of color tasks 27 ), memorizing a sequence of uncorrelated stimuli should decrease serial dependence in the course of an experimental session. Alternatively, if hard-wired mechanisms underlie serial dependence, we wouldn't expect any change.…”

“…The reasoning is that because similar stimuli usually elicit similar behavior, the brain would incorporate mechanisms to exploit these patterns 23 . Along these lines, a recent study has shown that systematic biases in color working memory change in the course of an experimental session to adapt to stimulus statistics 27 , arguing that systematic biases seen in delayed-estimation of color reflect real-world statistics. If similar adaptive plasticity operated for serial dependence in the time scale of the experimental session, we would expect to see a reduction of serial biases as one is exposed to a sequence of uncorrelated stimuli.…”

“…Some have further argued that serial dependence is of adaptive nature, changing its strength depending on the stimuli statistics 2,11,14,16,23 . Color, which is strongly correlated in natural scenes 24 , has never been studied with regard to its serial dependencies, possibly due to its strong systematic biases [25][26][27] . Similar to other perceptual biases for other visual features 28,29 , these systematic color biases adapt to stimulus statistics in the course of one experiment 27 .…”

“…Color, which is strongly correlated in natural scenes 24 , has never been studied with regard to its serial dependencies, possibly due to its strong systematic biases [25][26][27] . Similar to other perceptual biases for other visual features 28,29 , these systematic color biases adapt to stimulus statistics in the course of one experiment 27 . This suggests that typical perceptual bias adaptations occur in time scales of minutes to hours.…”