| Sparse Distributed Memory (SDM) is a content-addressable memory technique that relies on close memory items tending to be clustered together. One major disadvantage in the use of original SDM, as an associative memory, is that the input cue must be of sufficient length, typically a large percentage of the full stored word, in order to be able to retrieve associations from memory. In this work, I propose a variant of SDM, SDMSCue that is capable of handling substantially smaller input cues via space projection and successive reads. Results show superior recall for SDMSCue over original SDM.; I will also illustrate the use of SDM as an associative memory in the “conscious” software agent, CMattie, who is responsible for emailing weekly seminar announcements in an academic department. Interacting with seminar organizers via email in natural language, CMattie is to replace the secretary who normally handles such announcements. IDA, Intelligent Distribution Agent, uses SDM as well. IDA is intended to do the navy detailer job of assigning sailors to jobs according to their qualifications and preferences. SDM is the associative memory component in both complex architectures. CMattie and IDA implement global workspace theory. Global workspace theory is a psychological theory of consciousness and cognition by Baars. In this architecture, I will show how to use SDM, as the primary memory for the agent, to provide associations with incoming percepts, among other things. These include disambiguation of the percept by removing noise, correcting misspellings, and adding missing pieces of information. I also use SDM to retrieve behaviors and emotions associated with the percept. These associations are based on previous similar percepts, and their consequences that have been previously stored. SDM also possesses several key psychological features.; One final addition in my work is the use of GA (genetic algorithms) for the initialization of SDM instead of the random initialization widely used. This proves to be quite superior in terms of uniformity of the resulting semantic space. Results also show better recall rates of SDM initialized with GA over SDM initialized randomly. |
