Table 1.
AdEx parameters for tonic spiking used in this work.
Fig 1.
(a) The structure of a random linear genome in the population, encoding a network with three interneurons, three input nodes and one Output neuron.
A sequence of D elements (dendrites) followed by a sequence of AT elements (axon terminals) is considered one neuron. Each element has a type (I, O, D, AT), a sign (+, -) and (x, y) coordinates in 2D space. The strength of a connection between two elements is defined as an inverse function of the Euclidean distance between their coordinates. (b) Shows genetic elements (I, O, D, AT) in 2D space. (c) Network Topology of the obtained spiking neural network. Excitatory (inhibitory) connections are represented by red (blue) color.
Fig 2.
Networks recognising patterns of lengths 2 to 6. The first two networks in panels a and b were obtained with artificial evolution.
The blue (red) arrows represent inhibitory (excitatory) connections. The table below each network shows transitions between network states, represented by the number of active neurons. The topology is extended by hand to recognise patterns of lengths 4, 5, and 6 (c-e).
Fig 3.
The working mechanism of a network evolved in the presence of membrane potential noise.
(a) The states of the network and the corresponding activity of the neurons in the pruned network. (b) The minimal finite-state transducer that recognises ABC. (c) The activity of the network when a random stream of input signals A, B, and C is received. The different input levels represent input being received from different input channels; the voltage remains the same for all inputs. (d) The topology of the pruned evolved network.
Fig 4.
(a) Network activity of autapse-free network.
(b) Autapse-free network created by replacing N3 and N2 autaptic neurons in the three-signal network (see Fig 3 ) with mutual pairs of excitatory neurons.
Fig 5.
(a) The network states with corresponding active neurons.
(b) The handcrafted network for recognising a pattern of length six. (c) The behaviour of each neuron in the network when a random stream of input signals A, B, C, D, E, and F is received. The different input levels represent input being received from different input channels; the voltage remains the same for all inputs.
Fig 6.
Performance degradation of handcrafted networks with increasing pattern length.
The precision (left) and sensitivity (right) of the top 10 networks for each pattern size (3 to 6) are evaluated for a random sequence of length 1 million and all possible patterns of length , and n + 2.