Fourth, the names of the first three factors are confusing because they include dissimilar concepts.
In his theory of inference he distinguished between inference for oneself and inference for the other and laid down three criteria of a valid middle term (hetu i.e., that it should "cover" the minor premise (pakṣa), be present in the similar instances (sapakṣa), and be absent in dissimilar instances (vipakṣa).
This normalization is to balance the effect of similar instances and dissimilar instances.
Comparing across similar instances seemed to enhance memory for the individual items in a comparison more than across dissimilar instances.
Therefore, our objective function could be to minimize the distance between subgraph features of each pair of similar instances in each side view, and maximize the distance between dissimilar instances.
On similar and dissimilar instances (sapakṣa / vipakṣa) and the consequences of the various ways in which these notions are interpreted by Indo-Tibetan thinkers, see Tillemans (1999, ch. 5), Siderits (2003), Tillemans (2004b).
The third criterion of a valid reason the proof of the counter pervasion (vyatirekavyāpti / ldog khyab / ldog khyab), or the proof that the reason is completely absent from the dissimilar instances of the predicate (vipakṣa / mi mthun phyogs)—hold because there are no instances of the predicate which are not instances of the reason.
To investigate the effectiveness of the solution approach, two sets of instances are tested, including real-world instances and randomly generated instances.
Intuitively, Eq. (5) will minimize the distance between subgraph features of similar instance-pairs with (kappa ^{(p)}_{ij}ge mu ^{(p)}), while maximizing the distance between dissimilar instance-pairs with (kappa ^{(p)}_{ij}<mu ^{(p)}) in each view.
Feature's prediction power is recorded by its instances' distribution explicitly, which includes clear-discerned instances and blur-discerned instances.
Such a netlist needs to include instances of cell models.
