**9. Concluding remarks**

The article considers computational schemes of designing classifiers or prognostic models based on such a data set *C* (1), which consists of a small number *m* of highdimensional feature vectors **x**<sup>j</sup> (*m* < < *n*).

The concept of a complex layer composed of many linear prognostic models (88) built in low-dimensional feature subspaces is discussed in more detail. These models

(88) are built by using a small number *m* of collinear features *X*<sup>i</sup> belonging to the optimal feature clusters *Rl*\* (81). The optimal feature clusters *Rl*\* (81) are formed by the search for the largest margins δL1(**w***l*\*) (78) in the *L*<sup>1</sup> norm.

The averaged prognostic models *X*<sup>i</sup> 0 <sup>∧</sup> (89) are based on the layer of *L* parallel models *X*<sup>i</sup> <sup>0</sup>(*l*) (88). In line with the ergodic theory, averaging on a small number *m* of feature vectors **x**<sup>j</sup> has been replaced with averaging on *L* collinear clusters *Rl*\* (81) of features *X*i. Such averaging scheme should allow for a more stable extraction of general patterns from small samples of high-dimensional feature vectors **x**<sup>j</sup> (1) [11].
