Sentence examples for kernel similarities from inspiring English sources

The final kernel value is obtained as the sum of the similarities of the assigned atoms normalized by the self-similarities (i.e., the kernel similarities) of the respective molecules.

We have run numerous experiments for different values of k, and kept the value of k that visually yields the strongest bend in k-means potential and k-means intracluster kernel similarities.

This can be achieved by using a weighted average kernel similarity of a predicted molecule to the training data set.

Zhang et al. [4] proposed a kernel similarity modeling method for motion detection in complex and dynamic environments.

These methods rely on an implicit feature space that is only defined by the applied kernel similarity and with unknown dimensions.

Thus, in case of nonlinear SVMs, it is impossible to assess how a certain training set feature x ij contributes to the kernel similarity k(x i, x).

In this work, we propose three formulations of the applicability domains of kernel-based QSAR models, which rely only on the kernel similarity of the structures, and thus can be applied to assess the reliability of these models.

The Optimal Assignment kernel published by Fröhlich et al [46] computes the kernel similarity of two molecules by calculating the optimal bipartite graph matching of the atoms of the two molecules.

This density is identical to the average kernel similarity of x to the patterns x i from the training set X. Therefore, this approach is closely related to the similarity-based AD formulations [18, 19], with the difference that the formulation in this case ensures that the AD and the model are both defined in the same kernel-induced feature space.

In this work, three pure kernel-based descriptions of the applicability domain are introduced: the calculation of the subspace of the implicit feature space that contains most of the training patterns, which can be learned with a one-class support vector machine and two approaches to defining the domain boundary by a threshold for the (weighted) average kernel similarity.

Let us define k x as a kernel similarity function for drugs and k y as a kernel similarity function for target proteins.