Application of frequent episode discovery for analyzing multi-neuron spike train data

(i.e. no structure in the generating model). Episode counts from random data were compared with the episode counts from data with embedded patterns (as shown in Fig. 4).Noise: 1-6 data sets were generated by varying the random interconnections of the network of neurons and their firing rates.Pattern: 8-10 data sets were generated using logistic transfer function for neurons and with 10-node episodes embedded having conditional probability of connection = 0.8, 0.7 and 0.6 respectively.Pattern: 11-12 data sets were generated with linear transfer function and conditional probability of inter-connection in embedded episodes = 0.8 and 0.7 respectivelyIt can be seen that the frequencies of noise data fall to zero quickly with increasing size of episodes and are also orders of magnitude smaller than those for the data with patterns.Fig 4. Statistical Significance results for serial episodes