The geometry of domain-general performance monitoring in the human medial frontal cortex

Abstract

Flexibly adapting behavior to achieve a desired goal depends on the ability to monitor one’s own performance. A key open question is how performance monitoring can be both highly flexible to support multiple tasks and specialized to support specific tasks. We characterized performance monitoring representations by recording single neurons in the human medial frontal cortex (MFC). Subjects performed two tasks that involve three types of cognitive conflict. Neural population representations of conflict, error and control demand generalized across tasks and time while at the same time also encoding task specialization. This arose from a combination of single neurons whose responses were task-invariant and non-linearly mixed. Neurons encoding conflict ex-post served to iteratively update internal estimates of control demand as predicted by a Bayesian model. These findings reveal how the MFC representation of evaluative signals are both abstract and specific, suggesting a mechanism for computing and maintaining control demand estimates across trials and tasks.

Human acute microelectrode array recordings with broad cortical access, single-unit resolution and parallel behavioral monitoring

“…3). Per patient and recording session, this yield is similar to semichronic recordings in epilepsy patients (2 to 3 neurons per microwire bundle with up to 10 bundles implanted per patient (Fu et al, 2022;Kutter et al, 2018)). Acute DBS recordings from prefrontal cortex (10 to 20 neurons per participant (Jamali et al, 2019;Jamali et al, 2021)) or midbrain structures (fewer than 10 neurons per participant (Zaghloul et al, 2009;Zaghloul et al, 2012)) yield less.…”

“…Limited opportunities to directly access the human brain call for multidisciplinary collaborations that combine expertise in neuroscience and clinical medicine to invasively measure neuronal activity with single-unit resolution (Cash & Hochberg, 2015). This approach has been most fruitful in patients with medically intractable epilepsy implanted with microwire bundles (Fu et al, 2022; Kaminski et al, 2017; Kornblith et al, 2017; Kutter et al, 2018; Minxha et al, 2020; Rutishauser et al, 2010; Sheth et al, 2012) and in patients with movement disorders undergoing deep brain stimulation (DBS) (Jamali et al, 2019; Jamali et al, 2021; Zaghloul et al, 2009). Two crucial challenges persist, however, in the investigation of the cellular and circuit physiology of human brain functions.…”

Spontaneously emerging internal models of visual sequences combine abstract and event-specific information in the prefrontal cortex

“…At the same time the code was high-dimensional enough to allow the classification of diverse combinations of specific inputs. The same geometrical properties exist in the human medial frontal cortex for cognitive control, where the population code was both task-general for performance monitoring signals and task-specific in two paradigms, the Stroop task and a multisource interference task (Fu et al, 2022).…”

Revealing human sensitivity to a latent temporal structure of changes

“…Sequential activity of neuronal assemblies is one of principled neuronal operations that support higher level cognitive functions Eichenbaum (2014) ; Buzsáki and Llinás (2017) and allow humans to form complex spatio-temporal representation of our every day environment Frölich et al (2021) . Akin to grid cells known to support representation of both spatial and nonspatial task states Fu et al (2021) , time cells have been linked to temporal representation of state sequences critical for memory and decision-making Eichenbaum (2014) . Importantly, in spite of these fruitful experimental findings we have no clear computational understanding of how humans learn temporal structure in the service of successfully behavioural adaptation.…”