Learning and transfer of working memory gating policies

Bhandari, Apoorva; Badre, David

doi:10.1016/j.cognition.2017.12.001

Cited by 33 publications

(45 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Prior studies have reported similar RT dynamics as evidence of task automatization (Mohr et al, 2016). Our results indicate that this process should be generalized beyond S-R learning to control policy implementation (Bhandari and Badre, 2018). As such, novel S-R rules are not necessary to trigger adaptive processes indexed by early-trial RT dynamics.…”

Section: Discussionsupporting

confidence: 70%

“…As such, novel S-R rules are not necessary to trigger adaptive processes indexed by early-trial RT dynamics. Rather, S-R rule implementation is a special case of control policy implementation (Bhandari and Badre, 2018) i.e., task model assembly, (Duncan et al, 2008;Bhandari and Duncan, 2014) or control-context learning, (Chiu and Egner, 2019), required when task dynamics change.…”

Section: Discussionmentioning

confidence: 99%

“…This slower process might reflect formation of an abstract control policy participants can apply with consistency to a variety of trial structures. For example, consider an output-gating or "waiting" policy, wherein participants store all the cues, postponing all decisions to the end (Chatham et al, 2014;Bhandari and Badre, 2018). This policy uses working memory capacity inefficiently, but can be applied to a variety of task structures consistently.…”

Section: Discussionmentioning

confidence: 99%

“…As a result, we perform new tasks slowly and with errors, at least initially (Schneider and Shiffrin, 1977;Norman and Shallice, 1986). But with experience, we quickly settle on a cognitive routine for efficient task performance (Cole et al, 2010;Ruge and Wolfensteller, 2010;Wolfensteller and Ruge, 2012;Bhandari and Duncan, 2014;Bhandari and Badre, 2018). Such rapid 'task automatization' (Mohr et al, 2016) is observed across a wide variety of tasks.…”

Section: Introductionmentioning

confidence: 99%

“…Such rapid 'task automatization' (Mohr et al, 2016) is observed across a wide variety of tasks. This suggests that we can rapidly and flexibly implement an internal control policy that enables efficient coordination of cognitive processing (Bhandari and Duncan, 2014;Bhandari and Badre, 2018;Chiu and Egner, 2019). The neural systems that implement control policies remain understudied.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Fronto-parietal, cingulo-opercular and striatal contributions to learning and implementing control policies

Bhandari

Badrre

2020

Preprint

Self Cite

View full text Add to dashboard Cite

Efficient task performance requires co-ordination of internal cognitive processes by implementing control policies adapted to the dynamic structure of task demands. The cognitive and neural basis of control policy implementation remains poorly characterized, in part because it is typically confounded with implementing new stimulus-response rules. To disambiguate these processes, we asked participants to perform multiple novel variants of a working memory control task. Each variant had a unique, novel sequential trial structure, but all shared common stimulusresponse rules, enabling us to test control policy implementation separate from rule learning. Behaviorally, we found evidence for two adaptive processes tied to control policy implementation. One process was reflected in slower responses on the first trial with a novel sequential trial structure, followed by rapid speeding on subsequent trials. A second process was reflected in the diminishing size of the first trial cost as participants accommodated different variants of the task over many blocks. Using fMRI, we observed that the striatum and a cinguloopercular cortical network increased activity to the first trial, tracking the fast adjustment. This pattern of activity dissociated these regions from a fronto-parietal network including dorsolateral PFC, inferior frontal junction, inferior parietal sulcus, and rostrolateral PFC, which showed a slower decline in activity across trials, mirroring findings in rule implementation studies, but in the absence of rule implementation demands. Our results reveal two adaptive processes underlying the implementation of efficient, generalizable control policies, and suggest a broader account of the role of a cortico-striatal network in control policy implementation. Implementing Control Policies 3 Significance statementRapid adaptation to novel tasks is a hallmark of human behavior. Understanding how human brains achieve this is of critical importance in neuroscience. Here we broaden the scope of this problem, going beyond task rules to more broadly consider the cognitive control demands produced by novel task dynamics. We propose that humans rely on two adaptive processes to rapidly implement efficient, generalizable control policies as task dynamics change, even when task rules remain unchanged. One process unfolds rapidly and underlies efficient adaptation. A second process unfolds slowly with experience across task conditions and underlies generalization of control policies. Using fMRI, we identify cingulo-opercular cortex, fronto-parietal cortex and striatum as dissociable components of a cortico-striatal network that contribute to control implementation.

show abstract

Section: Discussionsupporting

confidence: 70%

Section: Discussionmentioning

confidence: 99%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Fronto-parietal, cingulo-opercular and striatal contributions to learning and implementing control policies

Bhandari

Badrre

2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

Working memory training restores aberrant brain activity in adult attention‐deficit hyperactivity disorder

Salmi

Soveri

Salmela

et al. 2020

Human Brain Mapping

View full text Add to dashboard Cite

The development of treatments for attention impairments is hampered by limited knowledge about the malleability of underlying neural functions. We conducted the first randomized controlled trial to determine the modulations of brain activity associated with working memory (WM) training in adults with attention‐deficit hyperactivity disorder (ADHD). At baseline, we assessed the aberrant functional brain activity in the n‐back WM task by comparing 44 adults with ADHD with 18 healthy controls using fMRI. Participants with ADHD were then randomized to train on an adaptive dual n‐back task or an active control task. We tested whether WM training elicits redistribution of brain activity as observed in healthy controls, and whether it might further restore aberrant activity related to ADHD. As expected, activity in areas of the default‐mode (DMN), salience (SN), sensory‐motor (SMN), frontoparietal (FPN), and subcortical (SCN) networks was decreased in participants with ADHD at pretest as compared with healthy controls, especially when the cognitive load was high. WM training modulated widespread FPN and SN areas, restoring some of the aberrant activity. Training effects were mainly observed as decreased brain activity during the trained task and increased activity during the untrained task, suggesting different neural mechanisms for trained and transfer tasks.

show abstract

A neural circuit model for a contextual association task inspired by recommender systems

et al. 2020

View full text Add to dashboard Cite

Behavioral data shows that humans and animals have the capacity to learn rules of associations applied to specific examples, and generalize these rules to a broad variety of contexts. This article focuses on neural circuit mechanisms to perform a context‐dependent association task that requires linking sensory stimuli to behavioral responses and generalizing to multiple other symmetrical contexts. The model uses neural gating units that regulate the pattern of physiological connectivity within the circuit. These neural gating units can be used in a learning framework that performs low‐rank matrix factorization analogous to recommender systems, allowing generalization with high accuracy to a wide range of additional symmetrical contexts. The neural gating units are trained with a biologically inspired framework involving traces of Hebbian modification that are updated based on the correct behavioral output of the network. This modeling demonstrates potential neural mechanisms for learning context‐dependent association rules and for the change in selectivity of neurophysiological responses in the hippocampus. The proposed computational model is evaluated using simulations of the learning process and the application of the model to new stimuli. Further, human subject behavioral experiments were performed and the results validate the key observation of a low‐rank synaptic matrix structure linking stimuli to responses.

show abstract

Learning and transfer of working memory gating policies

Cited by 33 publications

References 42 publications

Fronto-parietal, cingulo-opercular and striatal contributions to learning and implementing control policies

Fronto-parietal, cingulo-opercular and striatal contributions to learning and implementing control policies

Working memory training restores aberrant brain activity in adult attention‐deficit hyperactivity disorder

A neural circuit model for a contextual association task inspired by recommender systems

Contact Info

Product

Resources

About