Back

Early Emergence of Perceptual Biases in Working Memory

Serrano-Fernandez, L.; Romo, R.; Parga, N.

2025-04-27 neuroscience
10.1101/2025.04.26.650749 bioRxiv
Show abstract

Working memory (WM) is distributed across multiple cortical areas, suggesting that behaviors relying on WM arise from interactions between these regions. In a recent study, we found that during delayed comparison tasks, the first stimulus is not represented veridically in the prefrontal cortex (PFC), but instead is encoded in a systematically warped manner--biased toward the mean of the stimulus distribution. This neural distortion, which emerges already during the stimulus presentation and persists throughout the delay period, closely mirrors a contraction bias observed in behavior. Furthermore, the behavioral responses could be explained by a Bayesian observer model, in which the brain integrates prior expectations with noisy sensory inputs. These results suggest that the geometry of PFC neural trajectories embodies Bayesian estimates that underlie biased decisions. Here, we investigate whether the secondary somatosensory cortex (S2)--a lower-level sensory area also implicated in tactile WM--exhibits a similar encoding structure. Our analyses reveal that although WM-related signals in S2 are less robust than in PFC, the neural state space in S2 shares key geometric features with that of PFC, including a similarly warped representation of stimulus values. These findings suggest that perceptual biases may originate early in the cortical processing stream and are not exclusively shaped by higher-order associative areas. More broadly, our results support a distributed organization of WM, in which even sensory areas contribute to the formation of bias-prone representations that guide behavior.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.