Poster Presentation

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Making implicit assumptions of predictive processing explicit

Alejandro Tabas¹, Andrea Greve², Helen Blank³; ¹Basque Center on Cognition, Brain and Language, ²University of Cambridge, ³Ruhr University Bochum

Presenter: Alejandro Tabas

Predictive Processing (PP) is an umbrella term for mechanistic models of how the brain infers the sensory world. Although PP has been widely used to conceptualise experiments and interpret empirical results, their very success has led to a lack of conceptual rigour. Existing models incorporate numerous assumptions that are not always transparent to experimentalists, who often apply them in situations that violate the required assumptions. Here we formalise PP as an axiomatic framework building on two foundational assumptions: (1) the brain performs approximate Bayesian inference over the sensory input; (2) inference is approximately factorised along a representational hierarchy. Starting from these assumptions alone, we derive the computational elements required for perceptual inference, explicitly separating what is logically necessary, what is empirically established, and what remains unknown. We then apply the framework to two case studies, one theoretical and one experimental. First, we use the framework to study predictive coding under the Free Energy Principle, disentangling its implicit assumptions. Second, we map the empirical findings from classical oddball paradigms to the components of the framework, making clear what the empirical results show and what they do not. These two examples illustrate the power of the framework to clear up the conceptual fog. The framework offers three main contributions: a common language for dissecting prominent models and clarifying their empirically testable differences; a theoretical scaffold free from unnecessary assumptions to guide the interpretation of experimental findings without prematurely resolving unknowns; and a map of the most pressing open questions in the field. By replacing implicit assumptions with an explicit formal structure, this work aims to put the dialogue between theory and experiment in PP on a firmer ground.

Topic Area: Methods, Tools, Theory & Neural Coding