R. Auksztulewicz: Dynamic Predictions: Interactions Between Memory, Context, and Sensory Processing

Abstract
The brain is thought to continuously generate predictions based on past experience to optimize perception and behavior - a framework known as predictive processing. While this phenomenon is well supported across human and animal research, the extent to which contextual factors such as attention and task demands shape auditory predictions is less clear. In this talk, I present findings from a series of studies combining human and rodent neural recordings with computational modeling to uncover the mechanisms underlying sensory predictions and their interaction with other cognitive factors.

Our electrophysiological and neuroimaging results in humans show that predictive effects are not automatic but depend on context: both behavioral and neural data indicate that the relevance of predictions determines their impact. Computational modeling further distinguishes different forms of prediction - such as those related to stimulus content versus timing - linking them to separable neural mechanisms. Complementary studies in rodents demonstrate that auditory cortical activity encodes representations of both stimulus memory and predictions, suggesting that these mechanisms are evolutionarily conserved.

More recent work on associative learning reveals that prediction effects unfold dynamically across time. At short (peri-stimulus) timescales, valid predictions enhance neural decoding of visual categories, consistent with representational sharpening. At longer latencies, however, predictions suppress decoding, indicating a shift in processing dynamics. Across trials, early learning stages are characterized by dampening effects, whereas sharpening emerges more gradually over time.

Together, these findings highlight that predictive processing is not a uniform mechanism but is flexibly shaped by context. They underscore the dynamic nature of perception, in which past experience, current input, and future expectations are continuously integrated.

Bio
Dr. Ryszard Auksztulewicz is an Associate Professor at Maastricht University, where he studies predictive processing, learning, and memory using electrophysiology, neuroimaging, and computational modeling. He received his PhD from Humboldt-Universität zu Berlin and has held research positions at institutions including the University of Oxford, University College London, and City University of Hong Kong. His current work focuses on modelling individual variability in predictive processing and delineating memory from prediction signals in the brain.

Speaker's website

Zoom Meeting ID: 668 2663 6762
Password: BK-S2026