Neural representation of newly instructed rule identities during early implementation trials

Instructional load induces functional connectivity changes linked to task automaticity and mnemonic preference

Learning new rules rapidly and effectively via instructions is ubiquitous in our daily lives, yet the underlying cognitive and neural mechanisms are complex. Using functional magnetic resonance imaging we examined the effects of different instructional load conditions (4 vs. 10 stimulus-response rules) on functional couplings during rule implementation (always 4 rules). Focusing on connections of lateral prefrontal cortex (LPFC) regions, the results emphasized an opposing trend of load-related changes in LPFC-seeded couplings. On the one hand, during the low-load condition LPFC regions were more strongly coupled with cortical areas mostly assigned to networks such as the fronto-parietal network and the dorsal attention network. On the other hand, during the high-load condition, the same LPFC areas were more strongly coupled with default mode network areas. These results suggest differences in automated processing evoked by features of the instruction and an enduring response conflict mediated by lingering episodic long-term memory traces when instructional load exceeds working memory capacity limits. The ventrolateral prefrontal cortex (VLPFC) exhibited hemispherical differences regarding whole-brain coupling and practice-related dynamics. Left VLPFC connections showed a persistent load-related effect independent of practice and were associated with 'objective' learning success in overt behavioral performance, consistent with a role in mediating the enduring influence of the initially instructed task rules. Right VLPFC's connections, in turn, were more susceptible to practice-related effects, suggesting a more flexible role possibly related to ongoing rule updating processes throughout rule implementation.

Global connectivity fingerprints predict the domain generality of multiple-demand regions

A set of distributed cognitive control networks are known to contribute to diverse cognitive demands, yet it is unclear how these networks gain this domain-general capacity. We hypothesized that this capacity is largely due to the particular organization of the human brain's intrinsic network architecture. Specifically, we tested the possibility that each brain region's domain generality is reflected in its level of global (hub-like) intrinsic connectivity as well as its particular global connectivity pattern ("connectivity fingerprint"). Consistent with prior work, we found that cognitive control networks exhibited domain generality as they represented diverse task context information covering sensory, motor response, and logic rule domains. Supporting our hypothesis, we found that the level of global intrinsic connectivity (estimated with resting-state functional magnetic resonance imaging [fMRI]) was correlated with domain generality during tasks. Further, using a novel information fingerprint mapping approach, we found that each cognitive control region's unique rule response profile("information fingerprint") could be predicted based on its unique intrinsic connectivity fingerprint and the information content in regions outside cognitive control networks. Together, these results suggest that the human brain's intrinsic network architecture supports its ability to represent diverse cognitive task information largely via the location of multiple-demand regions within the brain's global network organization.

Instructing item-specific switch probability: expectations modulate stimulus-action priming

Both active response execution and passive listening to verbal codes (a form of instruction) in single prime trials lead to item-specific repetition priming effects when stimuli re-occur in single probe trials. This holds for task-specific classification (stimulus-classification, SC priming, e.g., apple-small) and action (stimulus-action, SA priming, e.g., apple-right key press). To address the influence of expectation on item-specific SC and SA associations, we tested if item-specific SC and SA priming effects were modulated by the instructed probability of re-encountering individual SC or SA mappings (25% vs. 75% instructed switch probability). Importantly, the experienced item-specific switch probability was always 50%. In Experiment 1 (N = 78), item-specific SA/SC switch  expectations affected SA, but not SC priming effects exclusively following active response execution. Experiment 2 (N = 40) was designed to emphasize SA priming by only including item-specific SC repetitions. This yielded stronger SA priming for 25% vs. 75% expected switch probability, both following response execution as in Experiment 1 and also following verbally coded SA associations. Together, these results suggest that SA priming effects, that is, the encoding and retrieval of SA associations, is modulated by item-specific switch expectation. Importantly, this expectation effect cannot be explained by item-specific associative learning mechanisms, as stimuli were primed and probed only once and participants experienced item-specific repetitions/switches equally often across stimuli independent of instructed switch probabilities. This corroborates and extends previous results by showing that SA priming effects are modulated by  expectation not only based on experienced item-specific switch probabilities, but also on mere instruction.

A One-Shot Shift from Explore to Exploit in Monkey Prefrontal Cortex

Much animal learning is slow, with cumulative changes in behavior driven by reward prediction errors. When the abstract structure of a problem is known, however, both animals and formal learning models can rapidly attach new items to their roles within this structure, sometimes in a single trial. Frontal cortex is likely to play a key role in this process. To examine information seeking and use in a known problem structure, we trained monkeys in an explore/exploit task, requiring the animal first to test objects for their association with reward, then, once rewarded objects were found, to reselect them on further trials for further rewards. Many cells in the frontal cortex showed an explore/exploit preference aligned with one-shot learning in the monkeys' behavior: the population switched from an explore state to an exploit state after a single trial of learning but partially maintained the explore state if an error indicated that learning had failed. Binary switch from explore to exploit was not explained by continuous changes linked to expectancy or prediction error. Explore/exploit preferences were independent for two stages of the trial: object selection and receipt of feedback. Within an established task structure, frontal activity may control the separate processes of explore and exploit, switching in one trial between the two.SIGNIFICANCE STATEMENT Much animal learning is slow, with cumulative changes in behavior driven by reward prediction errors. When the abstract structure a problem is known, however, both animals and formal learning models can rapidly attach new items to their roles within this structure. To address transitions in neural activity during one-shot learning, we trained monkeys in an explore/exploit task using familiar objects and a highly familiar task structure. When learning was rapid, many frontal neurons showed a binary, one-shot switch between explore and exploit. Within an established task structure, frontal activity may control the separate operations of exploring alternative objects to establish their current role, then exploiting this knowledge for further reward.

Instruction-based learning: A review

Humans are able to learn to implement novel rules from instructions rapidly, which is termed "instruction-based learning" (IBL). This remarkable ability is very important in our daily life in both learning individually or working as a team, and almost every psychology experiment starts with instructing participants. Many recent progresses have been made in IBL research both psychologically and neuroscientifically. In this review, we discuss the role of language in IBL, the importance of the first trial performance in IBL, why IBL should be considered as a goal-directed behavior, intelligence and IBL, cognitive flexibility and IBL, how behaviorally relevant information is processed in the lateral prefrontal cortex (LPFC), how the lateral frontal cortex (LFC) networks work as a functional hierarchy during IBL, and the cortical and subcortical contributions to IBL. Finally, we develop a neural working model for IBL and provide some sensible directions for future research.

Neural oscillations track the maintenance and proceduralization of novel instructions

Humans are capable of flexibly converting symbolic instructions into novel behaviors. Previous evidence and theoretical models suggest that the implementation of a novel instruction requires the reformatting of its declarative content into an action-oriented code optimized for the execution of the instructed behavior. While neuroimaging research focused on identifying the brain areas involved in such a process, the temporal and electrophysiological mechanisms remain poorly understood. These mechanisms, however, can provide information about the specific cognitive processes that characterize the proceduralization of information. In the present study, we recorded EEG activity while we asked participants to either simply maintain declaratively the content of novel S-R mappings or to proactively prepare for their implementation. By means of time-frequency analyses, we isolated the oscillatory features specific to the proceduralization of instructions. Implementation of the instructed mappings elicited stronger theta activity over frontal electrodes and suppression in mu and beta activity over central electrodes. On the contrary, activity in the alpha band, which has been shown to track the attentional deployment to task-relevant items, showed no differences between tasks. Together, these results support the idea that proceduralization of information is characterized by specific component processes such as orchestrating complex task settings and configuring the motor system that are not observed when instructions are held in a declarative format.

Frontoparietal action-oriented codes support novel instruction implementation

A key aspect of human cognitive flexibility concerns the ability to convert complex symbolic instructions into novel behaviors. Previous research proposes that this transformation is supported by two neurocognitive states: an initial declarative maintenance of task knowledge, and an implementation state necessary for optimal task execution. Furthermore, current models predict a crucial role of frontal and parietal brain regions in this process. However, whether declarative and procedural signals independently contribute to implementation remains unknown. We report the results of an fMRI experiment in which participants executed novel instructed stimulus-response associations. We then used a pattern-tracking procedure to quantify the contribution of format-unique signals during instruction implementation. This revealed independent procedural and declarative representations of novel S-Rs in frontoparietal areas, prior to execution. Critically, the degree of procedural activation predicted subsequent behavioral performance. Altogether, our results suggest an important contribution of frontoparietal regions to the neural architecture that regulates cognitive flexibility.

Temporal Learning Among Prefrontal and Striatal Ensembles

Behavioral flexibility requires the prefrontal cortex and striatum, but it is unclear if these structures play similar or distinct roles in adapting to novel circumstances. Here, we investigate neuronal ensembles in the medial frontal cortex (MFC) and the dorsomedial striatum (DMS) during one form of behavioral flexibility: learning a new temporal interval. We studied corticostriatal neuronal activity as rodents trained to respond after a 12-s fixed interval (FI12) learned to respond at a shorter 3-s fixed interval (FI3). On FI12 trials, we found that a key form of temporal processing—time-related ramping activity—decreased in the MFC but did not change in the DMS as animals learned to respond at a shorter interval. However, while MFC and DMS ramping was stable with successive days of two-interval performance, temporal decoding by DMS ensembles improved on FI3 trials. Finally, when comparing FI12 versus FI3 trials, we found that more DMS neurons than MFC neurons exhibited differential interval-related activity early in two-interval performance. These data suggest that the MFC and DMS play distinct roles during temporal learning and provide insight into corticostriatal circuits.