The memory model infers the recognition from these small portions of the entire memory. When input data are observed, edges based on the data are extracted with regard to the

edge configuration of the model. In the activation step, the extracted edge, Ei, and previously encoded edge, Em, are compared. To check the correspondence between two edges, the inclusion relation is MDV3100 Androgen Receptor inhibitor applied for a comparison measure. As a condition of the activation between two edges, at least one value should be matched, and no mismatched values should exist. The activation function can be represented as follows: δ(Ei,Em)={1,if  (#  of  matched  value>Nm, #  of  mismatched  value=0)0,otherwise. (3) If one value of an edge is missing because of different edge lengths, we do not count this case as a mismatched value. Figure 4 shows the success and failure of activation. Figure 4 Conditions of edge activation: (a) successful and (b) failed activations. The top and bottom rows are for the input edges and encoded memory edges, respectively. Arrows with a cross indicate mismatches of the edges between the input and memory. The secondary step of the recognition mechanism is judgment. To judge the familiarity,

an activation-based memory mechanism is involved. The model investigates whether the activated edges construct a fully connected links. After the edges are selected in the memory, the connected links are activated consecutively. If two adjacent edges are activated simultaneously by the input, the connected link is finally assigned as an activated link. If edges are activated and connected with each other in every

dimension of the network, the input data are judged as old (see (4)): ∏iδ(Ei,Em)=1. (4) If the activated edges are fully connected in the memory network, it means that the combination of edges was previously encoded. The reason for this is that all of the encoded instances make a closed link set in the network model. Figure 5 shows ring-type and line-type networks that have been judged as old or new. As shown in Figure 5, different edges are activated simultaneously. The number of closed loops changes according to the input data and network connectivity. However, the number of loops does not indicate certainty of the recognition judgment. The criterion is whether a fully connected link exists or not. Figure 5 Network diagrams built using activated edges and links. Top graphs ((a)–(d)) represent ring-type networks and bottom graphs ((e)–(h)) represent Carfilzomib line-type networks. Among the graphs, (a), (b), (e), and (f) contain a fully connected links, … 3.2.2. Performance Measure As a performance measure, we use a confusion matrix. In a hypernetwork, an old input is always judged as old if we assume that there is no removal of edges or links in the memory. This means that a false negative does not occur. Likewise, results judged as new are constantly made from new inputs. Our concern is false-positive cases, where a new input is judged as old.