An integrated learning rule for branch strength potentiation and STDP

R. Legenstein and W. Maass

Abstract:

Recent experimental data (Losonczy, Makara, and Magee, Nature 2008) show that not only the strength of synaptic efficacy is plastic, but also the coupling between dendritic branches and the soma (via dendritic spikes). More precisely, the strength of this coupling can be increased both through a coincidence of dendritic branch activations with action potential generation, and through a coincidence of branch activation with ACh. This effect has been called Branch Strength Potentiation (BSP). We show through theoretical analysis and computer simulations that the learning capability of single neurons is substantially increased if STDP is combined with BSP. More precisely, we show that a simple learning rule, based on a error-minimization principle, contains both BSP and STDP as special cases. The learning rule includes a homeostatic mechanism which acts locally at the site of the dendritic branch. The depression that was observed for post-before-pre pairings in standard STDP experiments is also observed in simulations of this learning rule. It can be explained by the combined effect of this local homeostatic mechanism and the backpropagating action potential. This powerful new learning rule endows single neurons with learning capabilities which were previously unattainable. For example, a single neuron acquires through this new learning rule the capability to solve a "binding problem". I.e., a single neuron can learn to respond to fire upon activation of presynaptic pools A and B, and also upon activation of presynaptic pools C and D, but NOT in response to concurrent activation of presynaptic pools A and C, or B and D. We also consider a variation of this learning rule where changes at synapses and branches are not only based on local activity, but also on a global reward signal that is indicated to the neuron by the concentration of a neuromodulatory signal such as ACh. We show that this biologically plausible learning rule for reward-based learning is much more efficient than previously proposed rules based on simple neuron models without nonlinear branches.



Reference: R. Legenstein and W. Maass. An integrated learning rule for branch strength potentiation and STDP. 39th Annual Conference of the Society for Neuroscience, Program 895.20, Poster HH36, 2009.