Home: Mini Project 3
Neural network for pattern recognition
Matlab's patternnet can generate networks with a user defined number of neurons and hidden layers. For example, 30 neurons in one hidden layer, or 10 each in three hidden layers.
Tetrode Data: hc-1.
There are four extracellular channels, one intracellular channel,
and one stimulus channel.
Waveform recognition of true spikes by the neural net is shown below. The first four channels are extracellular waveforms extracted per intracellular spike times (from "neuron A"). Not spikes (NS). NS 1 represents synthetic waveforms, and NS 2 represents detected spikes recorded on channel 2 that did not occur at the intracelluar spike times and, therefore, might be from one or more other neurons nearby. For this data, each waveform is 21 data points long, the neural net has one hidden layer with 30 neurons that use sigmoid activation functions, and the output layer uses a softmax function to output the probability that the waveform belongs to each of the six classifications.
d533101
