Title:
Minimal Predictive Systems
Abstract:
Evidence of prediction as the most important building block of cognition is mounting. Yet little is known about the requirements leading to the emergence of predictive abilities in any system. In this talk, I will first present my work on the epsilon-network, an artificial neural network for predictive coding. The epsilon-network adjusts its topology to the complexity of external inputs: it can be used as a measure of the predictability of the outside world.

I will then explain my proposal to implement a minimal predictive system, and the link between predictive systems and evolution: not only prediction is a function that can be evolved in biological systems, but evolution itself can be seen as a predictive function. Finding the definition of minimal predictive systems will tell us more about the conditions for the emergence of predictive abilities. By knowing the minimal conditions, we can identify existing predictive systems, study their evolutionary history and build totally new ones in unusual substrates.