It has been shown that the average mutual information is maximized by orthogonally-invariant random Welch bound equality (WBE) spreading sequences.
That is, when mutual information is maximized in the network model, the shape of the intraglomerular function is concave and the net LN input is inhibitory.
We found that mutual information was maximized when the intraglomerular transformation was concave (Fig. 3) and the LN input was inhibitory, which is consistent with previous experimental results [13], [31].
In this case, the mutual information was maximized when <img src="http://journals.plos.org/plosone/article/asset?id=info?doi/10.1371/journal.pone.0010644.e229.PNG" class= inline-graphic"/> was nearly 0, and the beneficial effect of LN input on the mutual information was significantly diminished.
How the LN input affects the PN responses can be visualized by comparing the PN response histogram at peak i with that at point n, where mutual information is maximized under the condition of no LN input ( <img src="http://journals.plos.org/plosone/article/asset?id=info?doi/10.1371/journal.pone.0010644.e149.PNG" class= inline-graphic"/> ) (Fig. 5 (A)).
By exploring possible nonlinear transformations produced by these two factors in our network model, we found that mutual information is maximized when a weak ORN input is preferentially amplified within a glomerulus and the net LN input to each glomerulus is inhibitory.
Similarity between both images was evaluated by normalized mutual information and maximized by Powell-Brent method.
In order to analyze the capabilities of an inference algorithm, instead of its employed statistical estimators, we follow [ 25] to obtain an optimal cut-off value I0 for the mutual information values by maximizing the F-score.
Using Monte Carlo simulations, we compare different methods (maximizing Youden index, maximizing mutual information, and logistic regression) for their ability to determine optimum binary cut-off thresholds for a ratio-scaled diagnostic test variable.
In order to discriminate the tastes in a mixture, the mutual information has to be maximized by maximizing the determinant of the covariance matrix.
The cost functions of PMMSE and PMMI are mean square error (to be minimized) and mutual information (to be maximized), respectively.
