Spike-based Expectation Maximization
Overview
Spike-based Expectation Maximization (SEM) is a theoretical framework that links the presence of spike-time dependent plasticity (STDP) and unsupervised learning. It has recently been shown that STDP, together with winner-take-all (WTA) inhibition is able to implement an expectation maximization step, and thus probabilistic inference. A neural network underlying these mechanism is able to discriminate its inputs, for example different poisson noise instances, or spike patterns representing hand-written digits. The model relies on additive plasticity, and positive weight changes must also depend exponentially on the current weight.
Developing institutions
TUG
Developers
Zeno Jonke
Results of the ESS workshop, Oct 2011
The model was presented at the Using the ESS + Neuromorphic hardware workshop in Dresden, Oct 4th, 2011 by Zeno Jonke. The slides of the presentation can be found here
- The weight dependence dor the STDP rule has been discussed and compared to current limitations of the hardware.
- Stochasticity of neurons difficult to implement, Concrete model suggested
- Relation between membrane potential and firing probability vastly different (exponential vs. almost linear)
