Hebbian learning with winner take all for spiking neural networks

Abstract

Learning methods for spiking neural networks are not as well developed as the traditional rate based networks, which widely use the back-propagation learning algorithm. We propose and implement an efficient Hebbian learning method with homeostasis for a network of spiking neurons. Similar to STDP, timing between spikes is used for synaptic modification. Homeostasis ensures that the synaptic weights are bounded and the learning is stable. The winner take all mechanism is also implemented to promote competitive learning among output neurons. We have implemented this method in a C++ object oriented code (called CSpike). We have tested the code on four images of Gabor filters and found bell-shaped tuning curves using 36 test set images of Gabor filters in different orientations. These bell-shapes curves are similar to those experimentally observed in the VI and MTN5 area of the mammalian brain.

Cell Microscopic Segmentation with Spiking Neuron Networks

“…This learning modifies the weights of the presynaptic neurons with the winning output [18]. This observation is in agreement with the fact that, in biological neural networks, different axonal connections will have different signal transmission delays [19].…”

Deep Blue Cannot Play Checkers: The Need for Generalized Intelligence for Mobile Robots

“…One would assume then that the mechanisms of subsymbolic generalization used by humans must somehow be different from the traditional generalizations in subsymbolic systems. However, perhaps the newer subsymbolic algorithms [8][9][10]34] will lead to more robust subsymbolic generalizations.…”

“…This would argue for a combination of approaches, with a subsumptive architecture being able to generalize over simple reactive tasks coupled with a symbolic system for generalization across more complex tasks. Some researchers have called for the combinations of Bayesian/probabilistic and connectionist approaches [34]. This would facilitate further understanding of the applicability of subsymbolic knowledge structures.…”

Pair-Associate Learning with Modulated Spike-Time Dependent Plasticity

“…In the second phase, the same mechanism is applied for reinforcement of the excitatory response groups but based on the number of spikes within 20 ms. In addition, winner-take-all strategy is implemented in both phases through biased random excitatory signals to the winner of response groups for each phase (adapted from [13]). The training performance is computed based on the percentage of number of correct response over number of trials averaged by 10 simulations (i.e.…”