Hippocampal spatio-temporal cognitive maps adaptively guide reward generalization

Abstract

The brain forms cognitive maps of relational knowledge, an organizing principle thought to underlie our ability to generalize and make inferences. However, how can a relevant map be selected in situations where a stimulus is embedded in multiple relational structures? Here, we find that both spatial and temporal cognitive maps influence generalization in a choice task, where spatial location determines reward magnitude. Mirroring behavior, the hippocampus not only builds a map of spatial relationships but also encodes temporal distances. As the task progresses, participants' choices become more influenced by spatial relationships, reflected in a strengthening of the spatial and a weakening of the temporal map. This change is driven by orbitofrontal cortex, which represents the evidence that an observed outcome is generated from the spatial rather than the temporal map and updates hippocampal representations accordingly. Taken together, this demonstrates how hippocampal cognitive maps are used and updated flexibly for inference.

Dynamic emergence of relational structure network in human brains

“…Reasoning the hidden relational structure from outside events is a crucial ability we human beings possess to help predict the future and make inferences (Balaguer et al, 2016; Garvert et al, 2021; Pudhiyidath et al, 2021). Here we demonstrate the behaviorally-related neural representations of two key aspects of relational knowledge – lower-order transition probability and higher-order community structure, which occur around 840 msec after image onset, before making a response.…”

“…Remarkably, humans could not only track transition probabilities, but also infer new abstract (higher-order) statistical structures from event sequences (Garvert et al, 2021;Mark et al, 2020). For example, humans tend to group events into clusters or hierarchical trees, which bias their subsequent reasonings (Balaguer et al, 2016;Pudhiyidath et al, 2021).…”

“…Events in the world are not isolated but are always interconnected according to some hidden structure, which we human beings are constantly pursuing (Pudhiyidath et al, 2021; Schapiro, Rogers, Cordova, Turk-Browne, & Botvinick, 2013). In fact, humans are gifted with tremendous abilities to discover the latent structure behind sequences of events, which is essential for rule inference, future prediction, and decision making (Balaguer, Spiers, Hassabis, & Summerfield, 2016; Garvert, Dolan, & Behrens, 2017; Garvert, Saanum, Schulz, Schuck, & Doeller, 2021; Pudhiyidath et al, 2021; Pudhiyidath, Roomea, Coughlina, Nguyena, & Preston, 2019). For example, after watching several scenes in a movie containing various character combinations, we could quickly form a structural network in mind that signifies connections among characters, such as families, lovers, friends, enemies, colleagues, etc., which are further attached to make a panoramic relational network.…”

“…Moreover, in addition to the one-dimensional transition chain, recent studies have investigated the learning of more complex networks, such as those with lattice, hexagonal, and community structures (Kahn, Karuza, Vettel, & Bassett, 2018; Kemp & Tenenbaum, 2008; Lynn, Kahn, Nyema, & Bassett, 2020; Mark, Moran, Parr, Kennerley, & Behrens, 2020; Schapiro et al, 2013). Remarkably, humans could not only track transition probabilities, but also infer new abstract (higher-order) statistical structures from event sequences (Garvert et al, 2021; Mark et al, 2020). For example, humans tend to group events into clusters or hierarchical trees, which bias their subsequent reasonings (Balaguer et al, 2016; Pudhiyidath et al, 2021).…”

Parallel cognitive maps for short-term statistical and long-term semantic relationships in the hippocampal formation

“…Abstracting and organizing relational information in this way facilitates flexible behavior, enabling generalization and inference. Beyond classical findings on the importance of cognitive maps for spatial navigation (e.g., Burgess, Maguire, & O’Keefe, 2002; Ekstrom & Ranganath, 2018; O’Keefe & Nadel, 1978), they are also thought to organize the relationships between objects (Constantinescu, O’Reilly, & Behrens, 2016; Garvert, Dolan, & Behrens, 2017; Garvert, Saanum, Schulz, Schuck, & Doeller, 2021; Morton, Schlichting, & Preston, 2020, Theves, Fernandez, & Doeller, 2019, 2020; Viganò, Rubino, Di Soccio, Buiatti, & Piazza, 2021), to represent temporal distances (Bellmund, Deuker, & Doeller, 2019; Bellmund, Deuker, Montijn, & Doeller, 2022; Burgess, Maguire, & O’Keefe, 2002; Schapiro, Kustner, & Turk-Browne, 2012; Solomon, Lega, Sperling, & Kahana, 2019), and to structure knowledge in the context of social cognition (Park, Miller, Nili, Ranganath, & Boorman, 2020; Son, Bhandari, & Feldmanhall, 2021; Tavares et al, 2015). While cognitive mapping is thus proposed to be a universal, domain-unspecific coding principle to systematically organize knowledge (Behrens et al, 2018; Bellmund, Gärdenfors, Moser, & Doeller, 2018; Stachenfeld, Botvinick, & Gershman, 2017), it is unclear how the brain handles stimuli embedded in multiple relational structures that are very distinct in terms of their mode and timescale of acquisition.…”