Relevance-based updates to contexts memorized during implicit statistical learning

Abstract

The implicit learning of unattended, task-irrelevant, statistical regularities that tempo-rally coincide with perceived successful task completion is a well-observed phenomenon. The cues to recall these memories and the ensuing guiding of attention to the reward-associated property are also driven by implicit cognitive processes. The ability to learn and recognize contexts, and to automatically guide attention to reward-associated features can provide obvious behavioral benefits. However, what if these environmental patterns are mere coincidences? Could these contextual memories be updated to reflect what the organism consciously believes is task-relevant as to dampen or remove their cueing effect? In this dissertation, I will propose a computational model of how relevance-based recon-solidation of contextual memories could occur. Leveraging the rich literatures of contextual cueing (Chun & Jiang, 1998), memory reconsolidation (Nader, 2016), and the Arcadia cognitive architecture (Bridewell & Bello, 2016), I’ve built a computational model of the proposed theory which performs a difficult visual-search task: learning, retrieving, updat-ing, and reconsolidating contextual memories.

The model is provided nearly identical stimulus as in an experiment from (Jiang & Leung, 2005). This experiment was chosen because of its focus on the role of attention in both implicit statistical learning and the retrieval of memories formed during implicit statistical learning. Provided nearly identical visual stimulus as human subjects, the model successfully simulates the contextual cueing effect (Chun & Jiang, 1998). Critically, the model also reproduces a key anomaly observed in (Jiang & Leung, 2005) and provides an alternative explanation appealing to relevance-based updates to the memory formed during implicit statistical learning versus Jiang and Leung’s appeal to a lack of learning due to associative blocking (Kamin, 1969). The modelling effort availed some interesting theoretical implications for implicit statistical learning regarding the interaction of reward, attention, and memory. The model’s results and implications are discussed at the end.