Minerva-DM

Model Description

Minerva-DM and Minerva 2 are akin to a single-layer neural network model, where the input corresponds to a pattern of features extracted from the environment and the output is a function of the contents of memory that are activated by the input. Both Minerva 2 and Minerva-DM assume an exemplar-based memory representation where each individual experience (i.e., episode) is represented by a distinct memory trace. Multiple experiences of similar events are therefore assumed to result in multiple, albeit similar, memory traces stored in memory. Because Minerva-DM preserved the representational and computational details inherent in Minerva 2, it can be used to simulate a variety of effects in the recognition memory literature, as well as the aforementioned phenomena in the frequency judgment and probability judgment literatures. Through the use of simulation methodology, Minerva-DM is able to make a priori predictions regarding the relationship between memory and judgment without the need to evoke specialized heuristic mechanisms.

Assumptions

Minerva-DM is based on the recognition memory model called Minerva 2. Minerva 2 makes two fundamental assumptions: (1) Memory consists of a database of instances that represent an individual's past experiences, and (2) recognition memory judgments are based on a global familiarity signal derived by matching a memory cue against all traces in memory simultaneously. Minerva 2 has been used successfully to model the influence of different types of experience on people's recognition and frequency judgments.

Minerva-DM extended Minerva 2′s capability by adding two additional assumptions. First, Minerva-DM assumes that memory traces can be partitioned into components that represent “hypotheses,” “data,” and “context.” Hypotheses correspond to events about which the participant is making a judgment, such as a disease hypothesis or a treatment hypothesis. The data component corresponds to the information on which the participant is making his or her assessment, such as the symptoms associated with diseases in past patients. The context component corresponds to environmental or task information available to the decision maker. The second additional assumption is that Minerva-DM assumes a conditional memory search process rather than a global memory search. The conditional memory search process involves first activating those memory traces in long-term memory that are consistent with the observable data (the presenting symptoms of the patient being diagnosed). The participant is then assumed to estimate the relative frequency of various disease hypotheses within the set of traces activated by the initial observable data. The relative frequencies are then normalized through a comparison process to derive a conditional probability judgment corresponding to the probability of the hypothesis in light of the data, that is, p(H|D).

...