Thu, Oct 14, 2010, 4:30 pm to 6:00 pm
We hypothesize that many large-scale biological networks organize themselves to operate in a dynamical regime where extensively many (in the thermodynamic sense) degrees of freedom ``poise themselves'' at the edge of a dynamical instability; we further hypothesize such dynamical criticality underlies the ability of the system to propagate information through the network, and to deploy different behaviors, system-wide, depending on context. We demonstrate dynamical laws that allow an artificial neural network to reach such state, i.e., to learn how to balance on a many-dimensional critical transition, and point out many consequences, such as on the size, spectrum, and spatial structure of fluctuations, that may help to identify such a state. We work out consequences for wave-like propagation of signals on abstract models of cortex. We apply these ideas to experimental data on ecocorticography array in humans and demonstrate signatures of both dynamical and statistical criticality in the recorded brain activity.