A computational model of replay-facilitated retroactive memory effects

Poster Session B - Sunday, April 14, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Emily T Cowan¹ (emily.cowan@temple.edu), Elizabeth A Horwath¹, Joseph E Dunsmoor², Vishnu P Murty¹; ¹Temple University, ²University of Texas, Austin

It is adaptive to prioritize the retention of salient information in long-term memory. However, we often only discover the value of given information through later, related experiences. Prior work has shown that exposure to salience cues (e.g., threat, reward) can retroactively enhance memory for conceptually related, but previously neutral content after a delay (Dunsmoor et al., 2015; Patil, Murty et al., 2017). The delay-dependent nature of this effect suggests it may be driven by consolidation-related mechanisms, such as replay, yet this hypothesis remains untested. Here, we adapted the Context Maintenance and Retrieval model (CMR, Polyn et al., 2009), to examine if reactivating memories during replay facilitates the retroactive spread of salient signals. To simulate prior behavioral work, inputs to the model were coded as two categories of stimuli which were encoded in two phases: a neutral “pre-conditioning” followed by a “conditioning” phase, in which one category was paired with a shock, simulated as an increase in learning rate. Memory was tested immediately or following a replay-filled delay, modeled as multiple cycles of iterative recall. Simulations using multiple variants of the model architecture showed that replay facilitated retroactive memory benefits when the model included context layers representing temporal context as well as a source context for stimulus category. In contrast, a model that included only temporal context did not lead to retroactive benefits. This work provides evidence for mechanisms by which the brain can tag and subsequently retain conceptually related content to facilitate adaptive functions of memory.

Topic Area: LONG-TERM MEMORY: Episodic