Schema-Driven, Space-Supported Random Accessible Memory Systems for Manipulation of Symbolic Working Memory

Schema-Driven, Space-Supported Random Accessible Memory Systems for Manipulation of Symbolic Working Memory Nader Noori (nnoori@usc.edu), Laurent Itti (itti@usc.edu) Department of Computer Science, University of Southern California Los Angeles, CA 90089 USA Abstract in the domain of cognitive psychology (CP) share is a meta- concept for their execution models which can be referred to as the dichotomy of process-storage. Applying this meta- concept to management of information for human working memory was one the most important contributions of Alan Baddeley and Graham Hitch to cognitive psychology which was originally presented in their seminal work (Baddeley & Hitch, 1974), and ever since has become the common denom- inator of all models of WM in CP. In this dichotomy which was inspired by Von Neumann’s architectural design for mod- ern digital computers (Von Neumann, 1982), the role of ex- ecution and processing is given to a central processing unit —namely the Central Executive (CE)—which controls the flow of information between and within storage slave units. However, a long debate over the nature of storage in CP com- munity (Jonides et al., 2008) has restricted elaborations on functional mechanisms of CE. We present an execution model for manipulation of working memory content during intellectual symbolic working memory tasks, which allows random access of WM content through a schema-operated sensory-motor spatial working memory. The core concept of this framework is binding symbolic items to spatial locations which are accessible via selective mecha- nisms of attention in space. An operational schema imple- ments basic WM management operations such as insertion, deletion and fetching through sequences of shifts in spatial at- tention towards registry locations. We apply the model to a se- rial recall task (both forward and backward orders). We show that the model provides a better fit to human data in backward recall compared to forward recall, which conforms with the evidence for leveraging spatial strategies for backward recall and phonological strategies for forward recall in normal sub- jects. We discuss additional possible implications of our model and its assumption of spatial organization of WM content and access through shifts of attention. Keywords: Memory Manipulation; Operational Schema; For- ward Recall; Reverse Recall; Computational Modelling; Intel- lectual Tasks; Working Memory. Introduction Cognitive psychologists use the term working memory (WM) to emphasize on the use of temporarily stored information in connection with cognitive tasks that involve processing in- formation (Baddeley, 1992). However, a review of the lit- erature shows that the information processing aspect of cog- nitive tasks mostly applies to and have been explored using intellectual tasks with symbols. Tasks such as random digit generation, forward and backward digit or word span, men- tal arithmetic, n-back recall, double counting and sorting are prevalent in the context of cognitive psychology to explore the ability for manipulation and maintenance of information in working memory (see Repov and Baddely’s review paper (Repovs & Baddeley, 2006)). Although the credit for popularizing the term working memory goes to cognitive psychologists, the concept of work- ing memory as the ability of temporarily storing information for the use in the upcoming task has been applied in other domains and to tasks that lack symbolic or intellectual fea- tures. For example, working memory which is of the interest in the perception community is related to maintenance and manipulation of information for sensory tasks such as visual search (Oh & Kim, 2004), or in action-perception domain for performing action routines (Arbib, 1987). However what distinguishes WM in different domains is beyond differences in particular instances of information and indeed is mostly related to their execution models: the func- tional principles for management or manipulation of informa- tion. In particular what different models of working memory The concept of CE in WM management did not prove as successful as its counterpart in Von Neumann’s proposal in achieving a working memory management schema which helps information processing. What distinguishes the central processing unit (CPU) in Von Neumann’s architecture from CE in Baddeley’s proposal is that the CPU had all mecha- nisms for control of storage units built in, while Baddeley and Hitch use the central executive as a metaphor for a cen- tral and powerful executive unit with no specific detail as to how CE controls slave storage units (Baddeley, 1992). As Baddeley himself has stated in several occasions because of this lack of specificity, CE has become the rag bag of unan- swered questions(Repovs & Baddeley, 2006) or a homuncu- lus (Repovs & Baddeley, 2006; Baddeley, 1996). What is known about the executive role of the central executive, for the most part, is postulated by Baddeley and colleagues. In- spired by Norman and Schallice’s idea of the Supervisiory Attentional System (SAS) (Norman & Shallice, 1986), Bad- deley has proposed that CE plays a role in controlling limited resources of executive attention (Repovs & Baddeley, 2006; Baddeley, 1996). However, adding the function of control- ling executive attentional resources has not been able to fill the void of a paradigm for an executive model for manipu- lation of information and to yield a model that explains how executive paradigm are encoded. To give an example of an alternative meta-concept for ma- nipulation of information we can refer to Arbib’s work on in- formation processing in perception-action loops. Arbib in his neuroethologic studies used the concept of information pro- cessing in a mechanistic fashion (Arbib, 1980) which was in-