Neurocomputational Dynamics of Sequence Learning

Konovalov, Arkady; Krajbich, Ian

doi:10.1016/j.neuron.2018.05.013

Cited by 38 publications

(32 citation statements)

References 73 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These aspects of behaviour were well accounted for by the normative two-system model , thereby showing that subjects use a common probabilistic currency to arbitrate between the different hypotheses, in line with previous accounts of Bayesian inference applied to discrete states 22,31,68 . We also explored alternative possibilities in the form of the non-commensurable two-system model which computes pseudo posterior probabilities of the non-random hypotheses independently of each other and, hence, cannot normatively compare them.…”

Section: Discussionsupporting

confidence: 83%

Rational arbitration between statistics and rules in human sequence processing

Maheu

Meyniel

Dehaene

2020

Preprint

View full text Add to dashboard Cite

5Detecting and learning temporal regularities is essential to accurately predict the future. 6 Past research indicates that humans are sensitive to two types of sequential regularities: 7 deterministic rules, which afford sure predictions, and statistical biases, which govern the 8 probabilities of individual items and their transitions. How does the human brain arbitrate 9 between those two types? We used finger tracking to continuously monitor the online 10 build-up of evidence, confidence, false alarms and changes-of-mind during sequence 11 learning. All these aspects of behaviour conformed tightly to a hierarchical Bayesian 12 inference model with distinct hypothesis spaces for statistics and rules, yet linked by a 13 single probabilistic currency. Alternative models based either on a single statistical 14 mechanism or on two non-commensurable systems were rejected. Our results indicate 15 that a hierarchical Bayesian inference mechanism, capable of operating over several 16 distinct hypothesis spaces, underlies the human capability to learn both statistics and 17 rules.

show abstract

Section: Discussionsupporting

confidence: 83%

Rational arbitration between statistics and rules in human sequence processing

Maheu

Meyniel

Dehaene

2020

Preprint

View full text Add to dashboard Cite

show abstract

“…For example, Bornstein and Daw (2012) employed a perceptual sequence learning task with an underlying probabilistic structure and found that the BG response to each stimulus was correlated with its forward entropy (the probability distribution of the possible next stimuli), whereas they suggested that hippocampal activity was related to “preparatory ‘prefetching’ of the anticipated next elements in the sequence” (specific predictions of sequence continuations; p. 1020). Using a similar sequence learning paradigm Konovalov and Krajbich (2018) found that the caudate response was related to prediction error; that is, violations of expectancy (which they align with Bornstein and Daw's forward entropy), whereas they suggest that the hippocampal response was related to pattern encoding (but not retrieval, which seems to contradict Bornstein and Daw). Wang, Shen, Tino, Welchman, and Kourtzi (2017) used a probabilistic system and the participants’ task was to make explicit predictions about sequence continuations with no feedback.…”

Section: Discussionmentioning

confidence: 90%

The Neuroscience of Implicit Learning

Williams

2020

Language Learning

View full text Add to dashboard Cite

Over the past decades, research employing artificial grammar, sequence learning, and statistical learning paradigms has flourished, not least because these methods appear to offer a window, albeit with a restricted view, on implicit learning processes underlying natural language learning. But these paradigms usually provide relatively little exposure, use meaningless stimuli, and do not even necessarily target natural language structures. So the question arises whether they engage the same brain regions as natural language. The aim of this review is to use data from brain imaging, brain stimulation, and the effects of brain damage to identify the main brain regions that show sensitivity to structural regularities in implicit learning paradigms and to consider their relationship to natural language processing and learning. This article is not a meta-analysis of studies in the field, nor will it advance a particular theoretical perspective, or systematically lay out an agenda for future research (though indications of outstanding questions and possible avenues for research emerge along the way). Rather it is an attempt to lay out the state of the art in the field for the nonspecialist, hoping to highlight, if nothing else, the complexities and inconsistencies in the evidence base, how these might be related to methodological variations, and to emphasize the problems inherent in making generalizations about localization of brain function on the basis of the evidence reviewed. With a view to the latter, the following section briefly lays out some of the ongoing debates over the function of core brain regions in relation to language processing and nonimplicit learning of natural languagelike systems.

show abstract

“…Neural network representations are often described as encoding latent information from a corpus of data (1)(2)(3)(4)(5)(6)(7)(8)(9). Similarly, the brain forms representations to help it overcome a formidable challenge: to organize episodes, tasks and behavior according to a priori unkown latent variables underlying the experienced sensory information.…”

Section: Introductionmentioning

confidence: 99%

Predictive learning as a network mechanism for extracting low-dimensional latent space representations

Recanatesi

Farrell

Lajoie

et al. 2018

Preprint

View full text Add to dashboard Cite

Neural networks have achieved many recent successes in solving sequential processing and planning tasks. Their success is often ascribed to the emergence of the task's low-dimensional latent structure in the network activity -i.e., in the learned neural representations. Similarly, biological neural circuits and in particular the hippocampus may produce representations that organize semantically related episodes. Here, we investigate the hypothesis that representations with low-dimensional latent structure, reflecting such semantic organization, result from learning to predict observations about the world. Specifically, we ask whether and when network mechanisms for sensory prediction coincide with those for extracting the underlying latent variables. Using a recurrent neural network model trained to predict a sequence of observations in a simulated spatial navigation task, we show that network dynamics exhibit low-dimensional but nonlinearly transformed representations of sensory inputs that capture the latent structure of the sensory environment. We quantify these results using nonlinear measures of intrinsic dimensionality which highlight the importance of the predictive aspect of neural representations, and provide mathematical arguments for when and why these representations emerge. We focus throughout on how our results can aid the analysis and interpretation of experimental data.

show abstract

Neurocomputational Dynamics of Sequence Learning

Cited by 38 publications

References 73 publications

Rational arbitration between statistics and rules in human sequence processing

Rational arbitration between statistics and rules in human sequence processing

The Neuroscience of Implicit Learning

Predictive learning as a network mechanism for extracting low-dimensional latent space representations

Contact Info

Product

Resources

About