Valmir C. Barbosa

2 papers in the library · publishing 2010-2016

Papers

Information integration from distributed threshold-based interactions

arXiv Preprint Archive June 27, 2016 Valmir C. Barbosa

A model of distributed units exchanging positive or negative messages based on thresholds abstracts key features of artificial intelligence and biological systems such as the brain. Information integration within a temporal window is quantified using total correlation, which measures information gain beyond individual units and relates to consciousness studies. Computational experiments explore how parameters (two probabilities and a threshold) affect total correlation as a function of window duration. Total correlation reaches significant fractions of its maximum. Semi-analytical results link message-traffic characteristics to window durations where total correlation peaks. Reinterpreting model parameters with cortical estimates yields optimal window durations aligned with time frames for conscious percept processing.

Full text Q-bio.nc Cond-mat.stat-mech

Network algorithmics and the emergence of information integration in cortical models

arXiv Preprint Archive December 27, 2010 Andre Nathan, Valmir C. Barbosa

A proposed alternative to integrated information theory (IIT) for studying consciousness uses information gain and total correlation to measure how efficiently a neural network generates information beyond what its parts produce independently. The framework applies to a previously introduced cortical model and compares it with random and deterministic network variants. Many instances of the cortical model integrate information more efficiently than most instances of the other models, suggesting its structure may support greater information integration.

Full text Q-bio.nc