Usually, k is smaller

than d and C is larger than d. Accordingly, dCk−d is less than 1, and thus the numbers of created edges and links PA-824 msds can more quickly converge than those of the instances. By using smaller number of edges, our proposed hypergraph structure can represent the entire instance combinations. 3.2. Inference Mechanism In summary, the proposed memory model is a layered hypergraph-based network. To operate as a recognition memory model, the model needs to facilitate both familiarity judgment and pattern completion. In this section, we deal with the judgment mechanism of hypergraph-based memory. In terms of the memory mechanism, there are two types of memory, activation-based and weight-based memory mechanisms [42]. A weight-based mechanism uses the

weights in the networks. A summation of all related weights is used to judge the classification of the input instance and categorize the output [43]. Previous global matching algorithms were built on the weight-based mechanism [31]. On the other hand, an activation-based mechanism adopts the shape of activation patterns as a judgment criterion. Previous researches on memory models have approached the functionality using a distinctive mechanism rather than mixing these two different mechanisms together [42, 44]. However, a hypernetwork has a particular connectivity in its structure and an individual weight for each connection, and thus the model represents two memory mechanisms together. As an activation-based mechanism, the model uses the shape of the activated edges and their connections. A weight-based mechanism enables measuring the intensity of the connections using the link weights. From the encoded memory, we describe the judgment mechanism of recognition memory through the two memory mechanisms. 3.2.1. Familiarity Judgment A constructed memory encodes all data into a hypergraph structure. When the new input data enter the memory, a recognition judgment begins. According to the completeness of the input data,

the process for the judgment is separated (see Figure 3). When an input Entinostat has no missing value, the result of the judgment is whether the input is old or new. On the other hand, a partial input to be judged requires distinctive processes related to the pattern completion. Inside the memory, the data commonly pass through the steps of edge sampling, activation, and finding fully activated connection. Figure 3 The recognition judgment procedure according to the type of input data. The upper arrows (dot lines) represent a process of familiarity judgment from complete input data. In contrast, the lower arrows (solid lines) show pattern completion from partial … The recognition judgment mechanism is divided into two steps: activation and judgment. The first, activation, is a step for finding the matched hyperedges from the input data.