RL underlying covert attentional selection

Top-down inputs from prefrontal cortex impact on sensory neurons, enhancing their selectivity to attended stimuli, while sensory processing of distractors is suppressed. However, what are the neuro-computational mechanisms that identify the behaviorally relevant information that is worth to bias? We recently introduced a reinforcement learning (RL) approach for the deployment of covert attentional selection. We tested model-free and model-based versions of RL: whereas model-based prioritizes attentional selection to task features that are systematically associated with reward, model-free considers all available features. Results prove that optimal task-set representations significantly improve predictive power, suggesting that primates benefit from model-based mechanisms.

Adaptive learning mechanism

Model-based reinforcement learning underlying covert attentional selection presents two limitations: i) it is unable to adapt to changes in the association of reward with sensory features that are not included in the learned model, and ii) it excludes the mechanism of learning by which subjects derive the proper task-set representation. The question remains then of how a prioritized task set can be learned to exploit the benefits of employing a model. With the aim to workaround model-based limitations and shed light on the underlying mechanisms that make model-based benefits possible, we propose the "adaptive learning" mechanism for flexible task-set representation. The mechanism is dynamically tuned according to the statistics of association between sensory features and reward outcome, flexibly adjusting its regime of operation between model-free and model-based systems. We tested the adaptive learning mechanism in a RL model that extended our previous analysis of monkey behavior to both, cued and uncued, versions of the same attention task. This RL model was able to transition from a naive starting point to an optimal task-set representation, and to flexibly adapt among optimal task-set representations upon changes in reward contingencies. The model achieved so by tracking a separate learning rate for each stimulus feature in the environment. The adaptive learning mechanism is a mechanistic candidate to support arbitrary prioritized model-based formation under general conditions of covert attentional selection, and regardless of whether attentional selection operates on cued or uncued tasks.

Functional classification of cells in primate PFC

Cortical microcircuits are composed of multiple cell classes that likely serve unique circuit operations, but the mapping is largely unknown, especially when it comes to function during actual goal-directed behavior in primates. One main problem in this quest is to reliably distinguish cell classes and characterize their firing and synchronization tuning in the circuit. Here, we surmount these difficulties and find reliably distinct properties of seven functional cell classes in the prefrontal cortex (PFC) of macaques engaged in a goal-directed attention task. We first delineate an unbiased bottom-up clustering protocol that identifies a functional hierarchy of four broad spiking (BS) and three narrow spiking (NS) cell classes based on their spike shape and how sparse, bursty, or regular they fire. We then show that these distinct classes of putative interneurons (NS cells) and pyramidal cells (BS cells) can be mapped onto four canonical functions of the underlying circuit: (i) Two putative pyramidal cell classes show sparse, bursty firing, and phase synchronize their spiking to theta and beta frequency bands. According to neuronal coding interpretations, these properties are optimally tuned for efficient input/output computations. ii) One NS and two BS cell classes show higher, regular firing. Circuit theories suggest these properties to be ideal for regulating the balance of excitation (BS) and inhibition (NS) and set it to a low level in the active PFC. iii) Two other NS classes fire irregularly and at low rates, and are dissociated by synchronization either to theta-, or to beta-frequencies of the local field potential. This narrow-band tuning to non-overlapping frequency ranges makes these cell classes resonant in different subnetworks. The resulting picture of the PFC fits with microcircuit models that propose a division of labor among cell classes in working memory and attention.

The tweaking principle for executive control

A hallmark of executive control is the brain's agility to shift between different tasks depending on the behavioral rule currently in play. In this project, we propose a 'tweaking hypothesis' for task switching: a weak rule signal provides a small bias that is dramatically amplified by reverberating attractor dynamics in neural circuits for stimulus categorization and action selection, leading to an all-or-none reconfiguration of sensory-motor mapping. Based on this principle, we developed a biologically-realistic model with multiple modules for attention switching. We found that the model quantitatively accounts for complex task switching behavior: switch cost, congruency effect and task-response interaction; as well as monkey's single-neuron activity associated with attention switching. The model predicts that category-selective neurons play a key role in resolving sensory-motor conflict.

Attentional processing in the neocortex

Selective attention is a fundamental cognitive function that uses top-down signals to orient and prioritize information processing in the brain. We have investigated a network model of spiking neurons composed of a reciprocally connected loop of two networks: a sensory area and a working memory circuit, source of the attentional top-down signal. A wide variety of physiological phenomena on sensory neurons induced by selective attention arises naturally in such a system. Our work demonstrates a neural circuit that instantiates the "feature-similarity gain modulation principle", according to which the attentional gain effect on sensory neuronal responses is a graded function of the difference between the attended feature and the preferred feature of the neuron, independent of the stimulus. Furthermore, our model identifies key circuit mechanisms that underlie feature-similarity gain modulation, multiplicative scaling of tuning curve, and biased competition.

Gamma-range synchronization in attention

In addition, we have investigated gamma synchrony associated with selective attention. Unlike models of coupled neural oscillators, we put the loop model in a "sparsely synchronized oscillation" regime, in which local field potential exhibits a population rhythm while single-cell firing is highly irregular, as observed experimentally. In this regime, top-down attentional inputs have a profound effect on oscillatory dynamics but they only slightly affect single-neuron spiking statistics. In addition, attentional synchrony modulations are highly selective, they occur only when there is a close match between neurons' preferred feature and the attended and stimulus features, which is a prediction to be tested. When inter-areal synchronization is abolished in the model, attentional modulations are reduced although in a moderated manner. Our model reconciles rate and synchrony effects, and suggests that inter-areal synchronization contributes to large-scale neuronal computation in the brain as well as to attention-specific enhancements of local oscillations.