Skip to main content
. 2013 Jun 6;7:10. doi: 10.3389/fnbot.2013.00010

Figure 3.

Figure 3

A large-scale network of ~4000 spiking neurons autonomously transitions between states reflecting a learned sequence. (A) Spike rastergram of all neurons in Networks A and B showing the population activity recorded over 700 ms. as the network spontaneously generated the learned sequence. Each spike is shown as a colored dot and each neuron is assigned a color to indicate the pattern to which it responds maximally. Networks A and B transitioned spontaneously between stable-states corresponding to three learned input patterns in the sequence as indicated by the three colors blue, green, and red associated with patterns four, five, and six, respectively. [The few magenta dots are associated with neurons responding best to another pattern, but which are also activated by pattern four (blue)]. The four neural populations in the network are labeled on the right of the diagram. Inh., Inhibitory; Exc., Excitatory. The labels T1 through T5 mark transition times referred to in the text. (B) After training, the coupled networks spontaneously generate a sequence of eight patterns in the correct order, and that sequence can be interrupted or shifted by presenting an external stimulus. Each row of the figure indicates, by brightness, the match score over time for one of the eight patterns that make up the sequence. The match score for pattern number X for example indicates how closely the neural population activity pattern in the Area A excitatory neurons matched the activity in the same population recorded when pattern X was presented to the network for the first time during training. White is a perfect match, while black indicates a complete mismatch. The network was trained from t = 0 to 32 s by stimulating the eight patterns in order. The times of stimulus presentation are indicated by black bars under the figure. The internally generated sequence is interruptible. When presented with one pattern in the sequence, pattern 6, for 1 s at t = 37 s and t = 45 s, the network activity immediately reflected the stimulus, and when the stimulus was removed the network generated the sequence from pattern 6 onward.