Unsupervised incremental STDP learning using forced firing of dormant or idle neurons

Spiking neural networks with hardware implementations of Spike Timing Dependent Plasticity (STDP) present a promising solution to energy efficient real-time machine learning. Online real-time learning, however, requires that new training information be considered by an already trained network without reinforcing previous data. Learning new information without severely altering previously learned data proves difficult in such networks. We present a spiking neural network capable of unsupervised incremental learning that exploits STDP to gracefully adapt the network under changing input conditions using a technique called forced firing. Our simulations validate this design with accuracy results comparable to networks that require data reinforcement.