Ventromedial prefrontal cortex compression during concept learning

Abstract

Prefrontal cortex (PFC) is thought to support the ability to focus on goal-relevant information by filtering out irrelevant information, a process akin to dimensionality reduction. Here, we test this dimensionality reduction hypothesis by relating a data-driven approach to characterizing the complexity of neural representation with a theoretically-supported computational model of learning. We find evidence of goal-directed dimensionality reduction within human ventromedial PFC during learning. Importantly, by using computational predictions of each participant's attentional strategies during learning, we find that that the degree of neural compression predicts an individual's ability to selectively attend to concept-specific information. These findings suggest a domain-general mechanism of learning through compression in ventromedial PFC.

Thalamocortical excitability modulation guides human perception under uncertainty

“…In particular, anterior and midline thalamic nuclei (such as the mediodorsal nucleus (MD), in which neuro-behavioral relations were maximal in our results) are implicated in establishing 43 , sustaining 44 , and switching 43 , 45 – 47 prefrontal rule representations depending on the active task context 43 . An intriguing possibility is that target uncertainty increases MD engagement to enable a more dynamic target selection among high-dimensional prefrontal feature representations 43 , 48 , 49 . Such proposal aligns with highest MD engagement during the integration of multiple cognitive demands 35 , 50 , and MD lesions specifically impairing performance for larger stimulus sets 51 and more complex tasks 52 .…”

The Costs and Benefits of Goal-Directed Attention in Deep Convolutional Neural Networks

“…Psychologists and neuroscientists have developed models that include goal-directed attention to explain behavioural (Bar 2006;Itti et al 1998;Love et al 2004;Miller and Cohen 2001;Nosofsky 1986;Plebanek and Sloutsky 2017;Wolfe 1994) and neuroimaging (Ahlheim and Love 2018;Mack et al 2016;Mack et al 2020) data. Algorithmically, goal-directed attention is often modelled as a set of weights that alter the importance of different psychological feature dimensions.…”

“…Using DCNNs as a starting point, we incorporate a simple goal-directed attention mechanism motivated by research in psychology and neuroscience (Nosofsky 1986;Kruschke 1992;Love et al 2004;Mack et al 2016;Mack et al 2020). We cast goal-directed attention as driven by a top-down signal that is, in ways, separate from the visual system (Fig.…”

Noise correlations for faster and more robust learning

“…Of particular importance is the assumption that signal-to-noise ratio of our populations is fixed, meaning that our manipulation of noise correlations can focus variance on specific dimensions without gaining or losing information. This assumption reflects conditions in which information is limited at the level of the inputs to the population, for instance due to noisy peripheral sensors (Beck et al, 2012;Kanitscheider et al, 2015;Mack, Preston, & Love, 2019). In such conditions, even with optimal encoding, population information saturates at an upper bound determined by the information available in the inputs to the population.…”

“…In particular, noise in task irrelevant dimensions might be considered in the same light that is often cast on suppression of task irrelevant dimensions by attentional mechanisms (Devauges & Sara, 1990;Lapiz & Morilak, 2006;Zanto & Gazzaley, 2009), in particular for purposes of accurate credit assignment (Akaishi et al, 2016;Joshi, Li, Kalwani, & Gold, 2016;Leong et al, 2017;Reimer et al, 2016). One possibility is that compressed low-dimensional task representations in higher-order decision regions (Joshi & Gold, n.d.;Mack et al, 2019;Vinck, Batista-Brito, Knoblich, & Cardin, 2015) may pass accumulated decision related information back to sensory regions in order to approximate Bayesian inference (Bondy et al, 2018;Bouret & Sara, 2005;Haefner et al, 2016;Lange et al, 2018). As task relevant features are learned, such a process would promote noise correlations between neurons coding those relevant features.…”