Figure 2

Neuromorphic implementation of SSA. Ten stimulations have been applied to the middle column to analyze our simulated-STD and the behaviour of a column in the network upon multiple stimulations. (a) represents the input coming to the column and (b) correspond to the output from the column. Synaptic depression has been previously shown to be key in the generation of PS and our model of spike frequency adaptation does simulate synaptic depression by the use of adapted inputs generated by the adaptive layer (a). This results in an increase of the transient MRF upon presentation of the first stimulus (orange rectangle) which does not occur for the next presentations (b). A PS appears upon the arrival of a stimulus and is represented by an increased MRF of the excitatory neuronal population in the column (red graph). The inhibitory population’s MRF (blue graph) increases only when a PS is generated and decreases through the course of the stimulation. Stimulus presentations (shown in gray) are of 100ms long followed by a pause of 900ms in between. (c) and (d) represents respectively the PS property and development in our network neuromorphic implementation. The time course of PS generation is depending on stimuli of varying amplitude and confirms the threshold for PS generation (c). Below an input stimulus of 5 spikes/sec, no PS were generated. The color scale represents the response strength. A PS appears upon presentation of input spikes and propagates through the network. The columns’ population activity (d) and the oscilloscope traces (see Supplementary data S2) demonstrate the propagation of spikes from the column where spikes were sent to the two adjacent columns. Following is the study of the dynamics of the network population in response to an oddball paradigm. (e) The firing rate of the network population of three columns (Deviant column (Dev), Recording column (Rec) and, Standard column (Std)) during an oddball paradigm show the appearance of SSA. PS were usually observed upon presentation of the deviant (stimuli presentation shown in blue (standard) and red (deviant) under the plot). (f) Upon presentation of the deviant, PS are propagating to one column from each side while the standard does not generate a PS, due to the simulated-STD. (g) The behaviour of both populations (excitatory and inhibitory) in the recording column has been studied and the activation of the inhibitory population occurs only while the deviant is presented. (h) The averaged responses of the recording column (spike count) following the different control OD protocols (as in Fig. 1e) were analysed, with the total spike counts normalized to the Deviant Alone condition. Each protocol consisted of 100 stimuli presentations, repeated 10 times, with an ISI = 1 second and the bar height represents the mean spike count with standard error of the mean. Similar results were obtained between the mean-field rate modelling and the neuromorphic implementation showing a tendency towards true deviance detection.