Computational and Neural Mechanisms Underlying Context Inference and Prediction

/ABSTRACT Cognition depends on context. The way we perceive stimuli, the predictions we make, and the actions we take all depend on the current situation. Considerable research has provided insight into how context affects neural processing, but relatively little is understood about how context itself is represented and learned. Here, we propose to combine computational models and electrophysiology in non-human primates to investigate the neural mechanisms that support context-dependent behavior. Our research builds on three recent theoretical models that use three different mechanisms for learning context representations and then using them to guide situationally-appropriate behaviors. These mechanisms include learning within prefrontal cortex, through the interactions between prefrontal cortex and basal ganglia, and through interactions between prefrontal cortex and hippocampus. To test the predictions of these models, we will simultaneously record neuronal activity from prefrontal cortex, hippocampus, and striatum of monkeys as they perform a context-dependent sequence prediction task. The proposed research has two primary aims: First, we aim to understand the structure of context representations in the brain. Monkeys will perform a sequential prediction task in which they must infer the context based on a cue and use it to predict subsequent stimuli. Each context will be associated with a unique sequence structure and designed in a way that allows us to understand the structure of the neural representation of context (as either compositional or conjunctive) and how this structure supports the generalization of knowledge between contexts. Recordings in prefrontal cortex, hippocampus, and striatum will test neural predictions from all three computational models about the nature of context representations in the brain. Second, we aim to understand how new contexts are learned. We will examine how different training regimes (e.g., blocked vs. interleaved contexts and transient vs. persistent cues) impact the formation and structure of context representations. Previous empirical and modeling work suggests that blocked training, while more difficult for standard neural networks, may benefit human learning by promoting compositional representations. Using neural recordings, we will test whether these findings extend to non-human primates and examine the role of prefrontal cortex, hippocampus, and striatum in learning under different training conditions. Overall, our research will provide insight into how the brain represents context and how these representations are shaped by learning experiences. This will refine our understanding of cognitive flexibility and lay the foundation for understanding, and addressing, disruptions in context-dependent processing associated with mental disorders including schizophrenia, obsessive-compulsive disorder, and anxiety. Project Number: 1R01MH141068-01A1 | Fiscal Year: 2026 | NIH Institute/Center: National Institute of Mental Health (NIMH) | Principal Investigator: Timothy Buschman (+1 co-PI) | Institution: PRINCETON UNIVERSITY, Princeton, NJ | Award Amount: $777,124 | Activity Code: R01 | Study Section: Human Complex Mental Function Study Section[HCMF] View on NIH RePORTER: https://reporter.nih.gov/project-details/11390316