Talk

Events

Events Search and Views Navigation

Event Views Navigation

Today

  • Fellow Lecture Neuroscience
    Genela Morris (Tel Aviv)
    Separate neurons for gains and losses: single-neuron recordings in the human brain

    NB 3/57
    Fellow Talk

    Human behavior is shaped by two opposing motivational drives: accumulation of reward and avoidance of harm. It is still unsettled whether, in the human brain, gains and losses are represented separately, enabling independent pursuit of both goals, or conjointly, by a collective pursuit on a common value scale. Human decision making is characterized by fundamental asymmetries between gain and loss processing, hinting at separate representation. Functional imaging studies nevertheless mostly favor a unified scale interpretation. We hypothesized that the poor spatial resolution of fMRI may account for these discrepancies: a voxel averages hundreds of thousands of neurons, so two intermingled populations with opposite preferences are indistinguishable from one bidirectional population. Here, we show that learning signals for gain and loss contexts are encoded by functionally separate neuronal populations that coexist within the same anatomical structures. We analyzed single-unit activity from 370 single neurons in prefrontal cortex, anterior cingulate cortex, temporal cortex, hippocampus and amygdala in 17 patients implanted with intracranial electrodes for epilepsy monitoring, while they performed a probabilistic choice task with separate gain and loss contexts. Trial-by-trial prediction errors were estimated from a reinforcement learning model fit to each participant’s choices. Gain and loss prediction errors were primarily represented by distinct, intermingled neuronal subpopulations, each selectively encoding context-specific prediction errors. A small subset of neurons encoded prediction errors in both contexts, frequently with opposite codes. Participants with greater segregation between the two codes showed a larger asymmetry in learning, with higher error and exploration rates in loss contexts than in gain contexts. Gain and loss errors therefore appear to be implemented through independent but anatomically interleaved neural channels, allowing independent processing pathways. This separation may offer insights into psychopathologies of biased valuation.

  • Philosophy
    Jakub Rudnicki (RUB)
    The semantics of event schema

    Online via Zoom

    Event schemas are knowledge structures pertaining to event types whose roles in our cognitive lives include guiding behavior, facilitating the encoding of new information, scaffolding memory retrieval processes (Ghosh & Gilboa 2014), and supporting the construction of episodic future thought representations (Addis 2020). They achieve these functions by associatively storing information about the typical ways in which events of a given type unfold, including the kinds of agents, objects, and subevents they typically involve, etc. Given that event schemas are, by definition, understood to result from experiences of multiple similar events that capture their general commonalities (Robin & Moscovitch 2017), they are naturally taken, in semantic terms, to have sets of events as their extensions. This, in turn, raises the question of how these extensions are determined. Given the aforementioned involvement of event schemas in the construction of episodic memory and episodic future thought representations, this question becomes especially pressing, as it may bear directly on the recently emerging debate over how reference is fixed in such representations (Openshaw & Michaelian 2024; Langland-Hassan 2024). The main aim of my talk is to offer an account of the semantics of event schemas. On my view, a given event schema refers to the set of events that it characterizes better than any other schema possessed by the organism. This notion of best characterization can be operationalized in terms of prediction error: a schema best characterizes an event if its application would yield the smallest prediction error among the organism’s available schemas. Additional support for this proposal comes from the leading model of event cognition, “Event Segmentation Theory” (Radvansky & Zacks 2011), according to which it is precisely prediction error exceeding a certain threshold that triggers the updating of the event model currently maintained in working memory.