Figure 3—figure supplement 1.

Model variations and optimization using EMOO.

(A–B) Bursts produced in simulations where the connections decay in space according to an exponential function (A), and where 20% of the connections were randomly eliminated (B). Left: Locations of bursts detected during a 30 min simulation. Colors indicate burst times and spot sizes indicate number of participating neurons. Right: Burst locations vs. burst time. Rectangle widths indicate the temporal extent of each burst. (C) Top: the change in bursting characteristics following 10% and 20% random synaptic pruning (grey) could be compensated for by re-tuning the model parameters (using evolutionary multi-objective optimization, EMOO, see below). Bottom: model parameters before and after re-tuning. (D) Schematic of EMOO (evolutionary multi-objective optimization). In each generation, a population of models are evaluated for their error scores on the three objectives and top-ranking models are selected. New models (‘children’) are produced by crossover and mutation from ‘parents’ and the extended population (selected top-ranking models and children) forms the next generation (see Materials and methods). (E) Model populations of the 2nd (left), 15th (middle), and 30th (right) EMOO generations, shown in the z-scored error space. Bold spots indicate the non-dominated set (first rank set), faint and fainter spots indicate the second and third rank sets, respectively. Red spot: baseline model; green spot: elbow solution used to initialise pattern search; blue spot: final optimized model that best matches biological bursting statistics. Parameter values indicated on the right.