Figure 3
From: Adaptive nodes enrich nonlinear cooperative learning beyond traditional adaptation by links
Dendritic learning as a self-controlled mechanism for oscillating weights, governed by the weak links. (a) A schema of a perceptron with seven inputs with weights and delays (w, τ, color coded). Changes in weights during the learning are defined in Fig. 2c, and the dynamics of the output is governed by leaky integrated-and-fire neuron (Methods). (b) A similar perceptron and delays as in a, but the output unit has three dendrites (color coded), red/green/orange connecting 3/2/2 input units, respectively, and with given initial dendritic weights, WD. (c 1 ) The initial seven weights in a are denoted (top, color coded). The input units are simultaneously stimulated at 10 Hz and the resulting dynamical evolution of the seven adaptive weights is presented. (c 2 ) Schematic presentation of the seven weights, ordered following their delays, with respect to \(\widetilde{{\rm{Th}}}\), the difference between the threshold and the current neuronal voltage, at three denoted timings (A–C) in c 1 . (d 1 ) Similar to c 1 where the same initial seven weights are now time-independent and the three dendritic weights in b are dynamically updated. (d 2 ), Similar to c 2 where the seven effective weights W*WD (color coded following their dendrites and ordered following their delays) are presented at five denoted timings (A–E) in d 1 . (e) and (f) present similar results as in c 1 and d 1 respectively, for a different set of seven initial weights.
