ABSTRACT
To best understand memory, we should be able to make reasonable predictions about how it works. This is best done using formal models of memory. Some models assume that memory retrieval involves a single process. This includes the threshold model, which is a simple model of memory. It also includes semantic network models that describe the structure of memory using associative relations among concepts and the spread of activation among them. Latent Semantic Analysis is another example, which uses large amounts of input data with many dimensions. Single process models also include multiple trace models (SAM and MINERVA 2), distributed storage models (TODAM and CHARM) that assume that memories are activated in a global matching process. Finally, PDP models use the structure of the nervous system as inspiration. Other models involve multiple processes, as with the generate-recognize model, the ACT model, and dual process models of memory that assume that there are at least two memory processes operating during retrieval: a familiarity component and a recollection component. Whatever the approach that is taken, the overarching goal is bring a greater degree of understanding and predictability to our approach to what goes right and wrong with human memory.
