Of Human Memory and Databases: Ardemia* The Relational Data Model and Management System as a Set Theory Based Modeling Architecture for Human Long Term Memory

This dissertation establishes the modelling and simulation architecture Ardemia*, which was designed for the experimental investigation of human memory and relational data. Ardemia supports modelling and simulation of human memory retrieval (remembering) and retrieval failure (forgetting) using a relational database management system (RDBMS). To this end Ardemia was designed to integrate essential properties of human long term memory (H-LTM) with Codd's relational model of data. Within this interdisciplinary framework, the primary research questions are: 1. Can the relational model of data be used to conceptualize the organization of HLTM? 2. Can a relational database management system be used to model human retrieval and forgetting? 3. Considering that the brain manages continuously streaming data with H-LTM capacities in the order of Petabytes over decades, can data management techniques be learned from H-LTM to address some of the challenges of big data?;This is the first study that brings together relational databases and H-LTM. To demonstrate a relationship between human Memory and databases, it would be a trivial task to a pick memory phenomenon of H-LTM and model it in a database environment. For example, one might hypothesize that cognitive schemas are a brain technique that can be useful for large databases to shrink their size or simplify searches. On the other hand, modelling H-LTM with a relational database system requires a deeper integration and the conceptual development of the commonalities between two inherently different systems. The gain from this extra effort is a controlled, empirical research environment and an experimental tool for (a) simulating memory phenomena and factors that affect human remembering and (b) the implications of these phenomena or manipulations for the storage and management of large data. So, instead of narrowly focusing on one memory phenomenon, we chose to investigate questions 1 and 2 and to lay the foundation for researching question 3. Following this introduction and the hypotheses overview, we proceed as follows:;In the foundation sections we present a review of the relational model of data and of its implementation. This is followed by a review of human memory and the identification of five dimensions and recurrent research themes that appear instrumental to our understanding and research of H-LTM.;In the conceptual integration section we begin by viewing H-LTM from a set theoretical perspective. This first level establishes the rationale of the project. This is followed by a high level mapping of the 5 H-LTM concepts to (a) the relational model of data and (b) current DBMS technologies.;In the implementation section the database begins with a visual overview of the task challenges; they are the development of Ardemia's ER model for H-LTM, the relational schema and SQL DDL code, and of the data generation and population based on a fictitious character, Mr. Polly.;In the experimentation section, the dimensions of H-LTM that Ardemia embodies are tested. The first set of simulations compares Ardemia's with Neural Network performance and establishes a set of benchmark queries. The second set of simulations compares the performance of humanized (heuristically degraded) databases to a control using the benchmark queries.;The results of the experiments indicate that Ardemia's associativity is at least as powerful as that of a neural network and that Ardemia's modeling and simulation approach is more parsimonious than the neural network implementation. Furthermore, Ardemia's sophisticated query capability supports the implementation of benchmark queries for use as experimental dependent variables as well as performance metrics. Last but not least, Ardemia is demonstrated as an experimental research environment to selectively model data (memory) impairments and generate data for comparison between models and existing empirical data.;In summary, Ardemia met the challenges posed by questions 1 and 2. We developed and tested a modelling and simulation architecture whose conceptual foundation integrates human long term memory and the relational model of data. Next is the investigation of question 3: As a simulation environment and modelling toolbox, Ardemia is ready for experimental computer and cognitive sciences to embark on a comparative investigation of artificial memory. It remains for us to improve our understanding of human long term memory, and to investigate its data reduction techniques and underlying storage model, which give rise to real time data decisions and action. This investigation lays the groundwork to use this knowledge to meet the challenges that are posed by managing and storing increasing volumes of data.;*The name "Ardemia" is pronounced ('Ar - Dee - Me - Ah') and the phonetic spelling of the acronym RDMA (Relational Data Memory Architecture).