Description

This app allows the user to interactively experiment with andvisualize the training of a single neuron (feature detector) in anunsupervised neural network. The user provides the online "trainingdata" via taking photos. Instructions: take photos and be patient-- even small neural networks are computationally intensive! Thenetwork converges on the first "principal component" or most oftenoccurring 2 x 2 feature pattern that multiple images share incommon. This network implements only local learning rules: 1. Aneuron is competitive: if it frequently does not fire, it decreasesits threshold to respond to stimuli with greater variance. 2. Aneuron is thrifty: if it frequently fires, it increases itsthreshold 3. When a neuron fires, it tunes its "weights" to move ina direction which will increase the likelihood of a similar patterntriggering it in the future. Example: if a post synaptic neuron wasresponding to a high amplitude signal, but now responds to a loweramplitude signal (due to threshold decrease), its weights will movetowards the lower amplitude signal. Network structure: 1. Inputlayer: 8 x 8 x 3 (3 color channels) 2. Convolutional layer: 2 x 2 x3 ( 1 neuron with 12 synapses) 3. Activation Map: 7 x 7 x 1 Note:Color channels are preprocessed to respond similarly regardless oflighting.