In order to record the continuous stream of autobiographical information that defines our unique personal history, our brains must rapidly construct durable episodic memories using only brief exposures to the patterns of activity generated by external stimuli. However, little is known about the computational strategies and biological underpinnings that make such “online” memory systems possible. Using both large-scale computer simulations and a simplified probability-based mathematical model, and taking into account recent evidence that dendrites may be the primary units where learning takes place, we investigated the relationship between dendritic arbor morphology and online memory capacity.
We found that recognition performance is maximized for dendrites of “medium” size (containing a few hundred synapses), whereas dendrites that are shorter (containing less than 100 synapses) or longer (containing more than 1000 synapses) suffer from severe capacity costs. We also investigated how the optimal dendrite size depends on various characteristics of the input patterns being learned, including (1) activation density, (2) noise, and (3) correlations. We found that increased density pushes the optimal morphology towards short dendrites, noise pushes the optimal morphology towards longer dendrites, and correlations reduce capacity and favor longer dendrites but may be counteracted by “duplication avoidance” in the wiring between dendrites and axons.
By helping to flesh out the causal chain that links the properties of neurons, dendrites, synapses, and patterns to online storage capacity, our results help elucidate not only the normal functioning of memory-related areas of the brain, but also which types of changes in neurons, their connections, and their plasticity rules that occur in aging, stress, neurological disorders, and mental retardation are likely to be most detrimental to memory function and why.
With: Bartlett Mel & Xundong Wu
A draft of this project is nearly complete and will be submitted soon, but in the meantime, here’s a poster from SfN 2011.