Model-based fMRI analysis of memory

Abstract

Recent advances in Model-based fMRI approaches enable researchers to investigate hypotheses about the time course and latent structure in data that were previously inaccessible. Cognitive models, especially when validated on multiple datasets, allow for additional constraints to be marshalled when interpreting neuroimaging data. Models can be related to BOLD response in a variety of ways, such as constraining the cognitive model by neural data, interpreting the neural data in light of behavioural fit, or simultaneously accounting for both neural and behavioural data. Using cognitive models as a lens on fMRI data is complementary to popular multivariate decoding and representational similarity analysis approaches. Indeed, these approaches can realise greater theoretical significance when situated within a model-based approach. HighlightsCognitive models can formalise theories to make assumptions and predictions clearer.Cognitive models offer additional constraints when interpreting the BOLD response.Models can go from behaviour to BOLD, vice versa, or address both simultaneously.Model-based approaches can incorporate pattern similarity and decoding approaches.

A controller-peripheral architecture and costly energy principle for learning

“…2B). Despite the system taking images as inputs, the controller's clustering solutions paralleled SUSTAIN's in terms of the modal number of clusters recruited (2,4,6,6,6,8 for Types I-VI respectively; for full results see Table S.1). Likewise, the controller's attention weights paralleled SUSTAIN's solution by selectively weighting the relevant stimulus dimensions (Fig.…”

“…We aim to address this gap by developing a general solution to the coordination problem and applying it to the domain of category learning, which requires the coordination of multiple cognitive processes related to attention, learning, object recognition, memory encoding and consolidation, and relies on coordinating multiple brain regions (e.g., ( 6, 7 )).…”

Toward a Neurocognitive Understanding of the Algorithms That Underlie Metamemory Judgments

“…Triangulating across areas of cognitive neuroscience provides strong leads on how regions involved in metamemory contribute to judgments, but strong tests of component-process accounts are still only beginning to emerge. One reason for the slow emergence of component process accounts in metamemory – compared to other areas of cognitive neuroscience (e.g., Forstmann & Wagenmakers, 2015) – is that metamemory lacks well-developed algorithmic models of latent cue utilization and decision processes that can be leveraged to provide quantifiable predictions for how these processes are engaged (e.g., Love, 2020). Such algorithmic models tend to depend heavily on ground-truth estimates of which cues are available in the task environment, which is a challenge in metacognition where many cues are likely internally generated and thus less observable.…”

Linking Models with Brain Measures

“…In this case, models are favoured to the extent that their internal state is decodable (Mack et al, 2013). (figure and discussion from Love, 2020b) MVPA decoding approaches apply a machine classifier to "mind read" from the BOLD response whether a participant, for example, is viewing a house or a face (Cetron et al, 2019). Although these are not psychological models, they can be used to make interesting behavioural predictions.…”

“…This work was supported by NIH Grant 1P01HD080679, Wellcome Trust Investigator Award WT106931MA, and Royal Society Wolfson Fellowship 183029 to B.C.L. Although mostly original, this paper draws on some previously published work (Love, 2020a(Love, , 2020bTurner et al, 2017). Thanks to Sebastian Bobadilla-Suarez for helpful comments on a previous draft.…”