Provisional Schedule:
08:30-09:00 registration with coffee
09:00-09:30 Randall O’Reilly: Introduction and overview
09:30-10:15 Floris de Lange
10:15-10:30 coffee break
10:30-11:15 Micha Heilbron
11:15-12:00 Fleur Zeldenrust
12:00-13:00 lunch (provided for registered participants)
13:00-13:45 Bernhard Englitz & Max Rensen
13:45-14:30 Timo van Kerkoerle
14:30-14:45 coffee break
14:45-15:30 Sabine Hunnius
15:30-16:30 Discussion
16:30-17:00 reception
The speakers in this workshop will discuss their empirical and theoretical work of relevance to the nature of cortical dynamics across time and between different layers and areas, in relation to ideas about the way that cortical learning might work, including predictive and other forms of learning. This includes work across the DCN, DCC, DCCN, and MPI, at different levels of analysis, from ion channels to human learning behavior.
The workshop is supported by the Radboud Excellence Initiative, with an award to visiting professor Randall O’Reilly, who has developed a computational model of learning in the neocortex based on the temporal derivative, i.e., a difference in neural activity over time, between the state of the network during the prediction, versus the state during the outcome (http://arxiv.org/abs/2606.08720). This contrasts with the classic Bayesian predictive coding framework (e.g., Rao & Ballard, 1999), which proposes that a sub-population of neurons directly represents the prediction error, by subtracting a top-down prediction from the bottom-up actual outcome. Thus, these two predictive learning frameworks predict qualitatively different types of cortical dynamics: one where the entire network is coherent and synergistic at any given point in time, versus one with more differentiated, specialized pathways.
Most of the available neural evidence is consistent with the coherent, synergistic encoding of information across all levels of the cortex. More generally, predictive learning must operate at multiple different time scales, not just different levels of hierarchical abstraction, which places further constraints on the nature of cortical dynamics that could support the processing and maintenance of predictions across these different time scales. These issues are relevant for the phenomena of statistical and sequence learning, and the metacognitive signals that might modulate learning.