Retrieving non-cued episodic memories by taking a cascading walk down memory lane(s)

Why are some memories brought to mind more easily than others? And how can we search our past experiences in a computationally tractable manner? In this work we propose that a process of cascading memory retrieval provides insight into both questions: memories are retrieved sequentially, whereby the retrieval of one memory provides the input for retrieving a subsequent memory. This process can be represented as sequential steps on a directed memory graph. Such a cascading walk through memory predicts two previously undocumented mnemonic phenomena: a predictable ordering in the sequence through which memories are retrieved, and that memories are less likely to be retrieved if the mnemonic path between a stimulus and the memory contains sub-paths with dead ends. We validate these phenomena through two experiments. As predicted, our findings reveal that the retrieval of past experiences is guided by the topology of the associative network. This cascading walk bridges between models of associative recall and stochastic sampling models, provides a mechanism to retrieve episodic memories in hippocampal (p)replay and mind-wandering, suggests that both episodic memories and semantic knowledge may be retrieved through a random-walk process, and provides a source mechanism for sequential sampling models.