Inferring latent contexts to guide behavior and memory

PROJECT SUMMARY Our everyday experiences are defined by environmental regularities. These regularities form stable contexts that enable us to make predictions about our environment (e.g., you are more likely to see your co-worker on a weekday than a weekend). In these cases, your context (day of the week) is inferred based on integrating

across experiences: in other words, it is latent. However, laboratory investigations of context have largely used observable manipulations (e.g., background color of the screen, task judgements, stimulus class). Latent contexts require cross-talk between mnemonic and perceptual systems to compare incoming evidence to

previously learned expectations. Here, we propose a novel experiment to determine the impact of latent contexts on brain activity and behavior. We will use item co-occurrences to instantiate latent contexts (e.g., seeing your co-worker, then the office cleaning staff, and then your boss is predictive of a weekday context).

We hypothesize that latent contexts will affect the way we represent information and perceive the environment. We will test this across 3 Aims using functional magnetic resonance imaging (fMRI), computational models, eye tracking, and behavioral assays. Across these aims we will interrogate the impact of latent states on the

brain’s representations of contexts (Aim 1) and items (Aim 2), and how these representations influence the allocation of attention (Aim 3). We predict that latent contexts require accumulation of evidence in regions that can integrate across experiences like hippocampus and posterior medial cortex. These regions will show shifts

in activity patterns that can be detected with machine learning models. Across the course of learning latent contexts, regions of the brain will re-shape how they represent associated items. Once item/context associations have been learned, patterns of activity will contain information about both the items themselves

and their contextual associations. We hypothesize that this shift in representations will predominately happen in regions that integrate across related items to track contexts, like the hippocampus and medial prefrontal cortex. However, representations in other regions that are attuned to perceptual information but not contextual

relationships, like visual cortex and perirhinal cortex, will not change across the course of learning. We also hypothesize that being in a latent context will facilitate predictions of upcoming information at the expense of detecting perceptual changes in the environment. Understanding latent contexts has important consequences

for how we characterize psychological and neural functioning in many psychiatric conditions like PTSD and anxiety in addition to the fundamental basic science questions about interactions between memory and perception that this proposal will answer.