查看文章

Availability of human memories for specific items is sensitive to frequency and recency. Sensitivity to these statistics has high adaptive value because they predict the likelihood of encountering the same items in the future (Anderson & Milson, 1989; Anderson & Schooler, 1991). This property of memory accounts for a range of cognitive phenomena and is implemented in various models (eg, Anderson & Milson, 1989; Petrov & Anderson, 2005). The ACT-R theory has been particularly successful and influential (Anderson & Lebiere, 1989; Anderson, Bothell, Byrne, Douglass, Lebiere, & Qin, 2004). The declarative memory in ACT-R is organized in chunks, and each chunk has a base-level activation (BLA) reflecting the frequency and recency of its use. The equation that updates these quantities has been “the most successful and frequently used part of the ACT-R theory”(Anderson et al., 2004, p. 1042, emphasis added). One serious practical drawback, however, is that it is very expensive to compute. ACT-R models have ground to a halt because of its complexity. This article describes an efficient approximation that scales up well and preserves all desirable properties of the exact equation.