Evaluating Approximations and Heuristic Measures of Integrated Information

• DOI: 10.20944/preprints201904.0077.v1
• Publication Date: 2019-04-08T05:46:47Z
• Authors: André Sevenius Ni...; Bjørn Erik Juel; William Marshall; Johan Frederik Storm
• Keywords: Approximations of π; Stochastic matrix; Matrix (chemical analysis); Information Theory

Integrated information theory (IIT) proposes a measure of integrated (Φ) to capture the level consciousness for physical system in given state. Unfortunately, calculating Φ itself is currently only possible very small model systems, and far from computable kinds systems typically associated with (brains). Here, we consider several proposed measures computational approximations, some which can be applied larger test if they correlate well Φ. While these approximations intuitions underlying IIT have had success practical applications, it has not been shown that actually quantify type specified by latest version IIT. In this study, evaluated heuristic measures, based on or clinical considerations, but rather how estimate values systems. To do this, simulated networks consisting 3–6 binary linear threshold nodes randomly connected excitatory inhibitory connections. For each system, then constructed system’s state transition probability matrix (TPM), as its (STM) over time all initial states. From matrices, calculated, Φ, differentiation, entropy, uniqueness, information. All were correlated dependent independent manner. Our findings suggest approximated closely using one more readily available (r > 0.95), without major reductions demands. Furthermore, strongly signal complexity (LZ, rs = 0.722), decoder (Φ*, 0.816), differentiation (D1, 0.827), (state independent). These could allow efficient estimation group level, accurate predictors low, high, it’s uncertain whether results extend other dynamics, stress importance aimed at being alternatives are minimum rigorously tested an environment where ground truth established.