From these two experiments, we evaluated the optimal human-like edge configuration for both a familiarity judgment MK 801 concentration and pattern completion. For these two tasks, fixed-order edges

showed a trade-off with the order sizes. Random-order edges with a random combination also showed a similar trend as fixed-order edges. In comparison, random-order edges showed regular ROC curves and a reasonable pattern completion performance regardless of the range of random orders. Hence, in the next experiment, we investigated the temporal properties of the proposed recognition memory model based on a random-order edge configuration. 4.3. Experiment 2: Investigate Temporal Encoding For the second experiment, we considered the properties of lifelong learning. We investigated the memory model in terms of the study duration and scale of encoded memory. Based on the first experiment, we evaluated whether the proposed recognition memory model resembles human performance through a comparison of the ROC curves. Later in the experiment, we investigated whether the scale of memory affects the familiarity judgment performance. Human familiarity capability was expected to be consistent regardless of the scale of information. However,

previous computational models on recognition memory have ignored this condition. Hence, we proved that our model is superior for lifelong learning by showing the performance consistency at different scales of encoded memory. 4.3.1. Temporal Encoding with Different Scale of Memory As the edge configuration, we assigned random-order edges with a range of (2, 5), which include as many various edge orders as possible. The dataset has about 7,000 instances in temporal order. In the same manner of evaluation, an instance is judged as old or new before the input instance is encoded. The performance was recalculated for every 1,000 instances that were encoded. The dataset was divided into seven subdataset.

In the first subdataset, there are no previously encoded data. In the second subdataset, 1,000 instances are tested in memory where previous Dacomitinib 1,000 instances were encoded. In the seventh subset, 6,000 instances have been encoded into memory, and the remaining instances are judged for evaluating the ROC curve. Figure 11 shows the ROC curves with different scales of memory. Overall, the shapes of the curves are constant except for the first and last sections. In the first section, the judgment performance was the highest. In contrast, the last section showed the lowest performance. However, the other middle sections were indistinguishable. Our proposed model produces a rather regular trend for temporal encoding. Figure 11 ROC curves with different scales of memory. Each curve is calculated using different scales of encoded memories and the same number of data to judge familiarity. 4.3.2.