Cognitive flexibility training for impact in real-world settings

Interacting with complex and dynamic environments challenges the brain’s ability to adapt to change. This key ability known as cognitive flexibility involves learning the structure of the environment, switching attention between features, dimensions and tasks, and adopting new rules in the face of uncertainty. Training cognitive flexibility has strong potential to improve adaptive behavior across the lifespan with impact in real-world settings (e.g. educational, clinical). Here, we review evidence on the role of cognitive training in improving executive functions and the factors that may enhance the effectiveness of training programs. We propose that personalized and adaptive training programs that focus on the multifaceted abilities comprising cognitive flexibility are key for promoting adaptive behavior and lifelong learning in real-world settings.


Introduction
Cognitive flexibility (CF), a multifaceted ability that is critical for successfully adapting to the needs and changes in dynamic environments [1], encompasses a range of skills, including attentional shifting, strategy updating, response to feedback, reversal learning, exploration, and task switching (Figure 1).In particular, task switching involves switching between different tasks or actions that are defined by specific stimulus-response associations, as demonstrated in the Trail Making Test or Task Set-Switching paradigm [2].Setshifting involves learning new rules with feedback and transitioning between cognitive sets that are defined by perceptual cues (e.g.color, shape), as tested by the Wisconsin Card Sorting task and the Intra-/Extra-Dimensional Set Shifting task.Reversal learning requires adjusting to changes in established reward-contingency associations.These skills collectively underpin our capacity to flexibly adapt to change and modify our behavior in dynamic environments [3].
This review highlights recent work on the role of training CF in promoting adaptive behavior and key considerations (Table 1) for CF training with potential impact in real-world settings (e.g.educational, clinical).

Cognitive flexibility: brain mechanisms
Studies investigating the neural substrates of CF highlight the role of fronto-striatal circuits [4][5][6].Animal studies have shown that the lateral prefrontal cortex is engaged in extradimensional shifts, while the orbitofrontal cortex (OFC) is key for reversal learning processes [7].Human neuroimaging studies have shown similar results; in particular, the ventromedial lateral prefrontal cortex has been implicated in extradimensional shifts (i.e. when a different stimulus dimension becomes task relevant), the lateral OFC in reversal learning (i.e. when a nontarget stimulus becomes target) following positive feedback, and the medial OFC in reversal learning following punishment [8].Furthermore, set-shifting has been shown to engage the

]] ]] ]]]]]]
dorsolateral prefrontal cortex (DLPFC), consistent with the cognitive demands of reconfiguring sets, while task switching has been shown to engage the anterior cingulate cortex and the pre-supplementary motor area (pre-SMA), which are known to be involved in executive control and managing multiple task demands.Taken together, these findings provide evidence for the nuanced and varied neural pathways underpinning the multiple aspects of CF, highlighting its complexity and the diverse neural architecture that supports flexible behavior.
Recent developmental studies propose that childhood and adolescence are critical periods for the maturation of brain regions that support CF [9], with evidence suggesting that this process may start as early as 9 months [10].This suggests that training of CF during early development may be critical for flexible behavior in adulthood.In particular, DLPFC and pre-SMA/SMA have been implicated in tasks requiring switching or attentional shifting in children [1].

Cognitive flexibility and learning under uncertainty
Recent work highlights the potential link between CF and probabilistic learning under uncertainty.The ability to shift between different strategies and mental representations -key aspects of CF -is critical when solving problems under uncertainty, for example, when navigating a new city or learning a new language.Recent  Designing training programs to improve CF.
1. Task-level considerations 1.1.Training higher-order control processes (i.e.abstract reasoning), rather than task-specific strategies (e.g.shifting) 1.2.Engaging neural and cognitive substrates that overlap with skills required for transfer 1.3.Personalized and adaptive training targeted to the participant's performance (e.g., enabling 'level' progression) 1.4.Training that challenges curiosity and inference abilities (e.g.changing the probability of getting higher scores, changing the task modalities, or implementing game-like elements) 2. Participant-level considerations 2.1.Sample size 2.2.Individual differences (e.g.SES, educational level, culture, personality, motivation, socioemotional traits) 2. studies employing a dynamic structure learning task (i.e.learning the underlying temporal structure of the environment without explicit feedback) demonstrate that individuals adapt to probabilistic changes in the statistics that govern the environment and exploit them to make predictions in novel settings [11][12][13].In particular, individuals change their decision strategy from matching (i.e.extracting exact sequence statistics) toward maximization (i.e.extracting the most probable outcome in a given context) to reduce uncertainty and improve their ability to learn predictive statistics in variable environments [12].Interestingly, individual decision strategies are shown to engage dissociable brain mechanisms [12][13][14][15]; that is, matching engages the visual corticostriatal pathway, whereas maximizing engages frontostriatal networks.Taken together, these studies suggest that structure learning under uncertainty facilitates flexible learning by switching between decision strategies that are supported by distinct brain circuits [16].Future studies are needed to investigate whether training interventions using dynamic structure learning paradigms may facilitate CF.

Cognitive training and real-world transfer
Most cognitive training programs have focused on skills related to memory, perceptual, and attentional control [17].As cognitive training aims to improve skills that are important for learning (e.g.problem-solving and flexible thinking) and real-life outcomes (e.g.educational attainment), its effectiveness is assessed based on three different types of generalization (i.e.transfer effects).Near-transfer involves improvement in tasks that assess the same processes as the training task but with different materials (e.g.stimuli) and/or task structure.Far-transfer involves improvement in tasks that require skills related to the cognitive domain trained by the intervention (e.g.training on a task involving spatial abilities that leads to improvements in verbal reasoning [18]).Finally, ecological transfer involves improvement in activities of daily life (e.g.critical thinking) and academic performance.
Executive functions have been associated with academic ability and educational attainment [19].In particular, meta-analysis studies have reported associations of working memory and CF with reading comprehension, reading efficiency, and mathematical skills [20,21].These findings have motivated the design of cognitive training programs for translation to educational settings [22,23].Despite associations between executive functions and real-world outcomes (e.g.academic attainment), cognitive training has mostly shown near-transfer on related domains [24,25] while evidence for far-or ecological-transfer effects [26] remains limited.Studies investigating the effect of cognitive training on academic abilities (e.g.math, reading, recalling of stories) have reported some beneficial effects of these interventions in typically and atypically (e.g.adult attention-deficit/hyperactivity disorder [ADHD] or dyscalculia) developing learners.Although most of the training programs have focused on working memory [19,27,28], studies focusing on inhibitory control and CF training also show beneficial effect on academic abilities (i.e.language and mathematical performance in school examinations, sentence comprehension in a standardized reading task [29-31]).Taken together, previous findings call for caution in interpreting cognitive training impact and highlight factors that may contribute to far-and ecological-transfer effects following training of executive functions (Table 1 Third, increasing engagement with cognitive training programs enhances motivation to learn and continue to train, enhancing their effectiveness.In particular, the perceived significance of the knowledge acquired, instructional support, interactive engagement (e.g. using computerized games), and teacher-student relationship for school-based cognitive training programs may determine the effectiveness of CF interventions in improving educational outcomes [32].For example, cognitive training has been shown to improve performance in healthy young adults (18-30 years) in the domain of sustained attention and concentration when using a game on an iPad to increase motivation during training [37].In healthy older adults (60-80+ years), cognitive functions across a range of domains could be trained using games on mobile phones [38].A meta-analysis of 16 studies with 1543 participants has confirmed that the domains of processing speed, verbal memory, and working memory can be successfully trained in older adults [39].Finally, reward based on task performance may enhance motivation with the potential caveat that offering excessive external rewards for an already internally rewarding behavior may reduce intrinsic motivation [40].Furthermore, recent work provides evidence that personalized and adaptive cognitive training using computerized games that enhance engagement and motivation to learn impacts cognitive performance in patient populations with neurological or psychiatric disorders.For example, cognitive training with a learning and memory game on an iPad was shown to improve performance not only in the training tasks but also in clinical measures for patients with stroke and patients with mild cognitive impairment and schizophrenia [41][42][43].This work provides promising insights for potential applications of CF training programs in clinical settings.For example, patients with obsessive-compulsive disorder (OCD) [6] or premanifest Huntington's disease (pre-HD) [5] show impairments in CF tasks.In OCD, impairments are linked to disrupted connectivity brain circuits known to be involved in CF, that is, between the caudate nucleus and ventrolateral prefrontal cortex, as well as the putamen and dorsolateral prefrontal cortex [6].Similarly, in pre-HD, cognitive impairments correlate with altered dorsolateral prefrontal cortex-caudate connectivity [5].These findings propose that training in CF tasks may enhance decision-making abilities and adaptive behavior in these patients.
Finally, methodological issues may obscure transfer effects in cognitive training studies.Some measures may be more sensitive than others; for example, cognitive training may result in improved academic potential, rather than academic performance; the latter is commonly used in cognitive training studies but may lack sensitivity to detect transfer effects [44].Furthermore, individual differences at baseline performance may obscure transfer effects if measured at group level [45].Therefore, more precise measures of individual performance may provide higher sensitivity in measuring fartransfer effects following cognitive training.

Training cognitive flexibility
Research on CF training is still in its infancy; yet, emerging research proposes that training CF may be effective in improving cognitive skills.Conditions that engage skills at the core of CF may facilitate improvement, that is, introducing variability in training protocols, uncertainty, and switching between task-relevant dimensions and across tasks.Furthermore, skills related to creativity that engage novel and divergent thinking (e.g.generation, remote association) may support CF.
The evidence for far-transfer effects following CF training is currently scarce but promising.Initial studies suggest notable improvements in task switching with transfer effects on different areas of executive functioning [46,47].Friedman and Miyake [48] have demonstrated through confirmatory factor analysis the potential impact of CF in real-world settings.Further studies suggest beneficial effects of CF training on academic abilities (i.e.language, performance in math tests, sentence comprehension in standardized reading tasks in children [49][50][51]).Training programs that target the diverse cognitive processes involved in CF may have stronger potential to show benefits that generalize to real-world settings.

Individual variability in cognitive training
Cognitive training outcomes are influenced by diverse factors, including age, socioeconomic status (SES), personality traits, and genetics.These factors influence the effectiveness of training programs, which varies greatly across individuals [45,52].For example, Karbach and Kray [47] found that generalization of task-set switching training was enhanced in adults and impaired in children when training involved novel task demands in each training session.Genetic predisposition is also found to modulate the effectiveness of cognitive training in CF.For example, Val/Val homozygous carriers of the COMT Val 158 Met polymorphism exhibited higher transfer in task switching than Met/-carriers, following video game training [53].Further investigation on individual differences in CF training is crucial for designing personalized cognitive training programs [25] and clarifying mixed results in generalization (i.e.near-vs far-transfer) of cognitive training [17].
Interrogating the processes underlying CF may enhance our understanding of individual differences in training outcomes.Computational modeling allows us to dissect the processes involved in CF by decomposing behavioral responses into process-specific modeling parameters [54].For example, modeling of structure learning tasks enables tracking individual decision strategies and how they change to help individuals learn, optimize task performance, and generalize [11].This is important for understanding how training individuals to change decision strategies may help optimize performance to solve complex tasks under uncertainty.

Conclusions
In sum, training that targets the multifaceted abilities that comprise CF and support adaptive behavior is critical for the effectiveness of cognitive training programs in real-world settings (e.g.educational, clinical).Furthermore, personalized adaptive designs that incorporate developmental, societal, and cultural factors and support enhanced engagement with training through games may increase the success of CF training and promote far-/ecological-transfer benefits to academic abilities [19,22,30

Figure 1 Current
Figure 1

3 .
Developmental stage (e.g.children, adolescents, young adults, older adults) 2.4.Neurodevelopmental conditions (e.g.dyslexia, dyscalculia, ADHD) 3. Context-level considerations 3.1.Delivery modality (i.e.computerized, online, in-person) 3.2.Delivery setting (i.e.individualized, classroom, home-based, community) 3.3.Cultural norms (i.e.training uses elements familiar to the individual's culture) 3.4.Engagement (i.e.computerized games to improve motivation) 4. Theoretical and measurement level considerations 4.1.Define the level of performance measurement (e.g.accuracy, speed, or computational-modeling index) 4.2.Define the set of skills being trained and how they relate to CF 4.3.Define the far-/ecological-transfer effects indicators and whether they are domain specific (e.g.math, reading comprehension) or domain general (e.g.intelligence) ).First, recent work suggests that cognitive training is enhanced if the content of the training task relates to the intervention target (e.g.generalization to academic skills) [32,33].For example, training children's CF skills by incorporating mathematical operations as task rules should benefit math performance and contribute to a better learning experience in mathematics, including mathematical flexibility [34].

Far transfer to language and math of a short software-based gaming intervention. Proc
]. Community's Seventh Framework Program (FP7/2007-2013) under agreement PITN-GA-2011-290011, and (2) the Cambridge-NTU Centre for Lifelong Learning and Individualised Cognition (CLIC), a project by the National Research Foundation, Prime Minister's Office, Singapore, under its Campus for Research Excellence and Technological Enterprise (NRF-CREATE SoL) Programme with the funding administered by the Cambridge Centre for Advanced Research and Education in Singapore Ltd. (CARES) and housed at the Centre for Research and Development in Learning (CRADLE@NTU).For the purpose of open access, the author has applied for a CC BY public copyright license to any author-accepted manuscript version arising from this submission.29.Goldin AP, et al.: This study describes a contextual approach to training executive function engagement for CF training to generalize to academic skills, rather than training executive functions directly that may improve educational outcomes.33.Gathercole SE, Dunning DL, Holmes J, Norris D: Working memory training involves learning new skills.J Mem Lang 2019, 105:19-42.34.Hong W, Star JR, Liu R-D, Jiang R, Fu X: A

systematic review of mathematical flexibility: concepts, measurements, and related research
This study shows that older adults improve from training with cognitive mobile games that increase engagement and motivation to learn.