We empirically tested a number of levels of regularization (from 10to50500000), but all values resulted in a degraded retinotopy (as quantified by the Spearman's rank correlation coefficient) over the default value of zero used here.
All sparse reconstruction algorithms involve a number of "knobs" such as threshold and regularization parameters that have to be properly tuned in order to converge to a unique solution.
A number of tools, such as L1-norm based regularization and its extensions elastic net and fused lasso, have been introduced to deal with this challenge.
Regularization is required here for a number of reasons: First, in order to help the minimizer combat the entrapment in a local minimum.
Results obtained by seven deep neural nets configurations over the seven bioactivity classes are shown in Fig. 3, here the effect of hyper-parameters (a) number of hidden layers, (b) number of neurons and (c) dropout regularization on the performance of DNN measured by MCC as evaluation metric are visualized averaged over the seven activity classes, while the rest parameters were kept fixed.
Here the analysis was focused on the hyper-parameters; (a) number of hidden layers, (b) learning rate, (c) number of neurons per hidden layer and (d) "dropout" regularization, while retaining the rest hyper-parameters fixed.
As a regularization method, the Tikhonov method has been used to solve ill-posed problems in a number of publications.
Our regularization is based upon the curvature constraint introduced by Fischer and Modersitzki (2004), since it exhibits a number of interesting properties.
In addition, its regularization term was designed to shrink a numbers of predictors to exactly zero.
* Please provide details on the classification (window size, number of classes, regularization factor, how many cycles?).
