The paper proposes a deep-learning approach to the recognition of melanoma images. It relies on the application of many different architectures of CNN combined in the form of an ensemble. The units of the highest efficiency are selected as the potential members of the ensemble. Different methods of arrangement of the ensemble members are studied and the limited number of the best units are included in the final form of an ensemble. The results of numerical experiments performed on the ISIC2017 database have shown the very high efficiency of the proposed ensemble system. The best accuracy in recognition of melanoma from non-melanoma cases obtained by the ensemble was 96.54% at AUC=0.9909, sensitivity 94.71%, and specificity 97.67%. These values are superior to the results presented for this ISIC2017 database.

The paper concerns the problem of minimization of the total potential energy of trusses subjected to static loads in the presence of prescribed displacements of selected supporting nodes. The positions of the internal (free) nodes are fixed and the supporting nodes are imposed, the member stiffnesses being design variables, while the truss volume represents the cost of the design. Due to the assumption of the stiffnesses being non-negative, the problem is reduced to a problem of optimization of structural topology. Upon eliminating all the design variables analytically the optimum design problem is eventually reduced to the two mutually dual problems expressed either in terms of member forces or in terms of displacements of free nodes. The problem setting concerning the case when the prescribed displacements of supports are the only loads applied (i.e. kinematic loads) assumes a particularly simple form. A specific numerical method of solving the stress-based auxiliary problem has been developed for the selected 2D and 3D optimal designs. The study is the first step towards topology optimization of trusses with distortions