Simulations of Efficient Experimental Design: A Minimum Entropy Approach

We apply an adaptive approach to optimal experimental design in the context of estimating the unknown parameters of a model of a neuron’s response. We define the optimal stimulus as the one which produces the most informative response. Our work presents an efficent implementation of an algorithm for choosing this stimulus on each trial. The algorithm requires no high-dimensional numerical optimizations or integrations even for high-dimensional stimulus and parameter spaces. Our simulation results show that model parameters can be estimated much more efficiently using this adaptive algorithm rather than random sampling. We also show that this adaptive approach leads to superior performance in the case that the model parameters are nonstationary, as would be expected in real experiments.