Therefore, when diverse types of edge orders are mixed into the m

Therefore, when diverse types of edge orders are mixed into the memory model, this tradeoff relationship can be resolved. 5.2. Meaning in Lifelong Learning The proposed model was developed to imitate the functionality of the human brain. During lifelong experience, humans occasionally Selumetinib ic50 become confused whether their current situation is familiar and may recognize new situations as old. This phenomenon happens when the person has already experienced a subset of the partial context. Since our hypergraph-based

memory model is also constructed by aggregating the subsets of the context, it shows a similar effect in the familiarity judgment. The purpose of recognition memory of lifelong experience is to recall and predict the user experience based on the previously encoded memory. The role of the recognition memory is determined according to the properties of the input data. When complete data are entered, the memory judges whether the event is old or new. For this function, the memory should perform the recall task well. On the other hand, partial data requires a different procedure based on the recognition memory. The partial data are generated into complete data through the encoded memory, and the completed data contain various

combinations including the exact original data. In terms of prediction, our recognition memory model suggests possible data from a partial input. It is assumed that the

memory has experienced the possible data before. Similar to memory with false alarms, humans can also be confused regarding their experiences. Furthermore, the new property of our computational model, that is, incremental recognition memory, can explain many unresolved phenomena in human behaviors. 5.3. Comparison with Other Models In order to build a computational recognition memory, previous researches on global matching algorithms [31] have also shown the human-like ROC performance on familiarity judgment. In comparison with the previous models, our proposed hypergraph-based memory model solves new issues related to recognition memory. First, the memory model is tractable to encode categorical data. The recognition memory is highly related to episodic memory and the dominant values of episodic memory are a sort Dacomitinib of categorical data. Hypernetworks encode the input data itself into the memory with special connection so that it can include any type of values as they are. Second, our model enables incremental learning without requiring the previously encoded data. In contrary, the global matching algorithms ignored the incremental learning issue and the structure of the models is fixed. Hence, if the model needs to be updated, it has to rebuild itself from all data. Third issue is that our memory model has high memory capacity. In lifelong experience, the detail values of contextual attributes are unlimited.

Leave a Reply

Your email address will not be published. Required fields are marked *

*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>