Abstract:

Predictive processing has emerged as a unifying principle for understanding cognitive processing in biological systems, yet its algorithmic and implementation-level foundations remain an active area of research. In this talk, I will present complementary lines of research investigating the informational architecture underlying predictive processing through a NeuroAI approach. First, I will discuss evidence for the neural mechanisms supporting predictive learning in biological systems, and how these computations are distributed across the cortical hierarchy. I will then show how deep neural networks can be used to test different predictive processing protocols and their computational properties. Finally, I will present how artificial systems can serve as naturalistic stimulus generators to probe and reveal fundamental properties of the informational architecture underlying predictive processing in biological systems.