Optimization of Neuronal Morphologies for Pattern Recognition
Speaker(s): Giseli de Sousa, University of HertfordshireAn important question related to the functional role of neuronal morphologies is why different types of neurons exhibit distinct dendritic trees. Previous studies have proposed that dendritic dimensions and branching structures are optimized to minimize the cost of transporting information from synapses to soma; other studies relate the dendritic topology, or the connectivity pattern of the dendritic segments, to the neuronal firing pattern. This study addresses the problem of how the dendritic structure and other morphological properties of the neuron can determine its pattern recognition performance.
Three techniques are used in this work for generating dendritic trees with different morphologies: exhaustively generating trees covering the whole search space; systematically generating tree samples; and evolving trees using an evolutionary algorithm. From these trees, I constructed full compartmental conductance-based models of neurons and assessed their performance by quantifying their ability to discriminate between learned and novel input patterns.
My results show that the morphology does have a considerable effect on pattern recognition performance. In fact, neurons with a small mean depth of their dendritic tree are the best pattern recognizers; the mean tree depth shows an anti-correlation with neuronal performance. I also demonstrate that the evolutionary algorithm could find effective morphologies for both passive models and models with active conductances. In fact, the evolved neurons performed at least five times better than the original hand-tuned neurons. In summary, the combination of morphological parameters plays a key role in determining the performance of neurons in the pattern recognition task and the right combination produces very well performing neurons.
Return to Seminars for 2011-2012