Context Representation, Interval Estimation, and the Parallax Effect in Human Memory

Abstract

Unlike most computers, the human brain doesn’t seem to contain a ticking quartz clock — yet, it’s still able to use a “sense of time” to retrieve memories and make sense of the progression of events in the past. Recent experimental findings have suggested that our ability to remember the order of events in time may rely, in part, on the brain storing memories alongside a snapshot of neural activation — an evolving mental ‘context’ — at the time of their creation. This thesis models these context values as vectors which, at every timestep, incorporate input which may take the form of pure noise or which may have a continuous directional signal. We model the process of making a single judgement in a sequential memory task as the result of a series of judgements on past events that are recalled alongside their context component. Finally, we compare the output of the judgements-on- traces process using different models of context and different inputs to each context model against performance by human participants on the same task. This allows us to explore whether certain patterns of error on sequential tasks could be due to ambiguity in a context-based ordering process in episodic memory.