Encoding in Balanced Networks: Revisiting Spike Patterns and Chaos in Stimulus-Driven Systems - Fig 8

<p>(<b>A</b>) Illustration of input similarity <i>ρ</i>_<i>same</i> for which the first <i>N</i>_<i>same</i> neurons receive identical direct inputs and . Also depicted is partial readout dimension <i>N</i>_<i>r</i> under two types of readout orderings: <i>ordered</i> and <i>random</i>, see text for details. (<b>B</b>) Scatter plot of classification performance <i>P</i> of surrogate population vs. network for a range of parameters <i>ρ</i>_<i>corr</i>, <i>ρ</i>_<i>same</i> and <i>N</i>_<i>r</i>, for both ordered and random readouts (same data as (E)). (<b>C</b>) <i>P</i> for network (black) and surrogate population (green) as a function of <i>N</i>_<i>r</i> with ordered readouts, for <i>ρ</i>_<i>corr</i> = 0, <i>ρ</i>_<i>same</i> = 0.5. Red dashed line indicates <i>N</i>_<i>r</i> = <i>N</i>_<i>same</i>. (<b>D</b>) <i>P</i> for network (black) and surrogate population (green) as a function of <i>N</i>_<i>r</i> with random readout, for <i>ρ</i>_<i>corr</i> = 0.9, <i>ρ</i>_<i>same</i> = 0. For (C) and (D), shaded areas show one one standard deviation (see sampling details below). (<b>E</b>) Color plot of <i>P</i> as a function of <i>ρ</i>_<i>same</i> and <i>ρ</i>_<i>corr</i> for <i>N</i>_<i>r</i> = <i>N</i>/4. Top row: ordered readouts (dashed line indicates <i>N</i>_<i>r</i> = <i>N</i>_<i>same</i>). Bottom row: random readouts. First column: network. Second column: surrogate population. For all <i>P</i> values shown, quantities were sampled over 20 repeated simulations where networks, inputs and Tempotron training were reprocessed. <i>N</i> = 500, <i>K</i> = 20.