scholarly journals Interpreting neural decoding models using grouped model reliance

2019 ◽  
Author(s):  
Simon Valentin ◽  
Maximilian Harkotte ◽  
Tzvetan Popov
Keyword(s):  
Machine Learning ◽  
Working Memory ◽  
Memory Load ◽  
Memory Task ◽  
Machine Learning Algorithms ◽  
Support Vector ◽  
Neural Decoding ◽  
Working Memory Load ◽  
Eeg Data

AbstractThe application of machine learning algorithms for decoding psychological constructs based on neural data is becoming increasingly popular. However, there is a need for methods that allow to interpret trained decoding models, as a step towards bridging the gap between theory-driven cognitive neuroscience and data-driven decoding approaches. The present study demonstrates grouped model reliance as a model-agnostic permutation-based approach to this problem. Grouped model reliance indicates the extent to which a trained model relies on conceptually related groups of variables, such as frequency bands or regions of interest in electroencephalographic (EEG) data. As a case study to demonstrate the method, random forest and support vector machine models were trained on within-participant single-trial EEG data from a Sternberg working memory task. Participants were asked to memorize a sequence of digits (0–9), varying randomly in length between one, four and seven digits, where EEG recordings for working memory load estimation were taken from a 3-second retention interval. Present results confirm previous findings in so far, as both random forest and support vector machine models relied on alpha-band activity in most subjects. However, as revealed by further analyses, patterns in frequency and particularly topography varied considerably between individuals, pointing to more pronounced inter-individual differences than reported previously.Author summaryModern machine learning algorithms currently receive considerable attention for their predictive power in neural decoding applications. However, there is a need for methods that make such predictive models interpretable. In the present work, we address the problem of assessing which aspects of the input data a trained model relies upon to make predictions. We demonstrate the use of grouped model-reliance as a generally applicable method for interpreting neural decoding models. Illustrating the method on a case study, we employed an experimental design in which a comparably small number of participants (10) completed a large number of trials (972) over multiple electroencephalography (EEG) recording sessions from a Sternberg working memory task. Trained decoding models consistently relied on alpha frequency activity, which is in line with existing research on the relationship between neural oscillations and working memory. However, our analyses also indicate large inter-individual variability with respect to the relation between activity patterns and working memory load in frequency and topography. Taken together, we argue that grouped model reliance provides a useful tool to better understand the workings of (sometimes otherwise black-box) decoding models.

scholarly journals Unity and diversity in working memory load: Evidence for the separability of the executive functions updating and inhibition using machine learning

2018 ◽  
Author(s):  
Tanja Krumpe ◽  
Christian Scharinger ◽  
Wolfgang Rosenstiel ◽  
Peter Gerjets ◽  
Martin Spüler
Keyword(s):  
Machine Learning ◽  
Working Memory ◽  
Executive Functions ◽  
Memory Load ◽  
Brain Computer Interface ◽  
Theoretical Models ◽  
Computer Interface ◽  
Support Vector ◽  
Learning Approaches ◽  
Eeg Data

AbstractObjectiveAccording to current theoretical models of working memory (WM), executive functions (EFs) like updating, inhibition and shifting play an important role in WM functioning. The models state that EFs highly correlate with each other but also have some individual variance which makes them separable processes. Since this theory has mostly been substantiated with behavioral data like reaction time and the ability to execute a task correctly, the aim of this paper is to find evidence for diversity (unique properties) of the EFs updating and inhibition in neural correlates of EEG data by means of using brain-computer interface (BCI) methods as a research tool. To highlight the benefit of this approach we compare this new methodology to classical analysis approaches.MethodsAn existing study has been reinvestigated by applying neurophysiological analysis in combination with support vector machine (SVM) classification on recorded electroenzephalography (EEG) data to determine the separability and variety of the two EFs updating and inhibition on a single trial basis.ResultsThe SVM weights reveal a set of distinct features as well as a set of shared features for the two EFs updating and inhibition in the theta and the alpha band power.SignificanceIn this paper we find evidence that correlates for unity and diversity of EFs can be found in neurophysiological data. Machine learning approaches reveal shared but also distinct properties for the EFs. This study shows that using methods from brain-computer interface (BCI) research, like machine learning, as a tool for the validation of psychological models and theoretical constructs is a new approach that is highly versatile and could lead to many new insights.

Abnormal prefrontal activity subserving attentional control of emotion in remitted depressed patients during a working memory task with emotional distracters

2011 ◽  
Vol 42 (1) ◽  
pp. 29-40 ◽  
Author(s):  
R. Kerestes ◽  
C. D. Ladouceur ◽  
S. Meda ◽  
P. J. Nathan ◽  
H. P. Blumberg ◽  
...  
Keyword(s):  
Working Memory ◽  
Attentional Control ◽  
Memory Load ◽  
Memory Task ◽  
Region Of Interest ◽  
Blood Oxygen Level Dependent ◽  
Emotional Information ◽  
Working Memory Load ◽  
Depressed Patients ◽  
The Right

BackgroundPatients with major depressive disorder (MDD) show deficits in processing of facial emotions that persist beyond recovery and cessation of treatment. Abnormalities in neural areas supporting attentional control and emotion processing in remitted depressed (rMDD) patients suggests that there may be enduring, trait-like abnormalities in key neural circuits at the interface of cognition and emotion, but this issue has not been studied systematically.MethodNineteen euthymic, medication-free rMDD patients (mean age 33.6 years; mean duration of illness 34 months) and 20 age- and gender-matched healthy controls (HC; mean age 35.8 years) performed the Emotional Face N-Back (EFNBACK) task, a working memory task with emotional distracter stimuli. We used blood oxygen level-dependent (BOLD) functional magnetic resonance imaging (fMRI) to measure neural activity in the dorsolateral (DLPFC) and ventrolateral prefrontal cortex (VLPFC), orbitofrontal cortex (OFC), ventral striatum and amygdala, using a region of interest (ROI) approach in SPM2.ResultsrMDD patients exhibited significantly greater activity relative to HC in the left DLPFC [Brodmann area (BA) 9/46] in response to negative emotional distracters during high working memory load. By contrast, rMDD patients exhibited significantly lower activity in the right DLPFC and left VLPFC compared to HC in response to positive emotional distracters during high working memory load. These effects occurred during accurate task performance.ConclusionsRemitted depressed patients may continue to exhibit attentional biases toward negative emotional information, reflected by greater recruitment of prefrontal regions implicated in attentional control in the context of negative emotional information.

scholarly journals Contralateral Delay Activity Tracks Fluctuations in Working Memory Performance

2018 ◽  
Vol 30 (9) ◽  
pp. 1229-1240 ◽  
Author(s):  
Kirsten C. S. Adam ◽  
Matthew K. Robison ◽  
Edward K. Vogel
Keyword(s):  
Working Memory ◽  
Individual Differences ◽  
Memory Load ◽  
Memory Performance ◽  
Memory Task ◽  
Memory Storage ◽  
Alpha Power ◽  
Working Memory Load ◽  
Contralateral Delay Activity ◽  
Trial Performance

Neural measures of working memory storage, such as the contralateral delay activity (CDA), are powerful tools in working memory research. CDA amplitude is sensitive to working memory load, reaches an asymptote at known behavioral limits, and predicts individual differences in capacity. An open question, however, is whether neural measures of load also track trial-by-trial fluctuations in performance. Here, we used a whole-report working memory task to test the relationship between CDA amplitude and working memory performance. If working memory failures are due to decision-based errors and retrieval failures, CDA amplitude would not differentiate good and poor performance trials when load is held constant. If failures arise during storage, then CDA amplitude should track both working memory load and trial-by-trial performance. As expected, CDA amplitude tracked load (Experiment 1), reaching an asymptote at three items. In Experiment 2, we tracked fluctuations in trial-by-trial performance. CDA amplitude was larger (more negative) for high-performance trials compared with low-performance trials, suggesting that fluctuations in performance were related to the successful storage of items. During working memory failures, participants oriented their attention to the correct side of the screen (lateralized P1) and maintained covert attention to the correct side during the delay period (lateralized alpha power suppression). Despite the preservation of attentional orienting, we found impairments consistent with an executive attention theory of individual differences in working memory capacity; fluctuations in executive control (indexed by pretrial frontal theta power) may be to blame for storage failures.

Effects of a secondary task and working memory load on multisensory hand position

2012 ◽  
Vol 25 (0) ◽  
pp. 58
Author(s):  
Katrina Quinn ◽  
Francia Acosta-Saltos ◽  
Jan W. de Fockert ◽  
Charles Spence ◽  
Andrew J. Bremner
Keyword(s):  
Working Memory ◽  
Visual Information ◽  
Secondary Task ◽  
Memory Load ◽  
Memory Task ◽  
Hand Position ◽  
Working Memory Load ◽  
The Individual ◽  
Task Conditions ◽  
Relative Weighting

Information about where our hands are arises from different sensory modalities; chiefly proprioception and vision. These inputs differ in variability from situation to situation (or task to task). According to the idea of ‘optimal integration’, the information provided by different sources is combined in proportion to their relative reliabilities, thus maximizing the reliability of the combined estimate. It is uncertain whether optimal multisensory integration of multisensory contributions to limb position requires executive resources. If so, then it should be possible to observe effects of secondary task performance and/or working memory load (WML) on the relative weighting of the senses under conditions of crossmodal sensory conflict. Alternatively, an integrated signal may be affected by upstream influences of WML or a secondary task on the reliabilities of the individual sensory inputs. We examine these possibilities in two experiments which examine effects of WML on reaching tasks in which bisensory visual-proprioceptive (Exp. 1), and unisensory proprioceptive (Exp. 2) cues to hand position are provided. WML increased visual capture under conditions of visual-proprioceptive conflict, regardless of the direction of visual-proprioceptive conflict, and the degree of load imposed. This indicates that task-switching (rather than WML load) leads to an increased reliance on visual information regardless of its task-specific reliability (Exp. 1). This could not be explained due to an increase in the variability of proprioception under secondary working memory task conditions (Exp. 2). We conclude that executive resources are involved in the relative weighting of visual and proprioceptive cues to hand position.

scholarly journals Working Memory and the Suppression of Reflexive Saccades

2002 ◽  
Vol 14 (1) ◽  
pp. 95-103 ◽  
Author(s):  
Jason P. Mitchell ◽  
C. Neil Macrae ◽  
Iain D. Gilchrist
Keyword(s):  
Working Memory ◽  
Memory Load ◽  
Critical Role ◽  
Memory Task ◽  
Behavioral Responses ◽  
Error Rates ◽  
Antisaccade Task ◽  
Working Memory Load ◽  
Reflexive Saccades ◽  
Back Task

Conscious behavioral intentions can frequently fail under conditions of attentional depletion. In attempting to trace the cognitive origin of this effect, we hypothesized that failures of action control—specifically, oculomotor movement—can result from the imposition of fronto-executive load. To evaluate this prediction, participants performed an antisaccade task while simultaneously completing a working-memory task that is known to make variable demands on prefrontal processes (n-back task, see Jonides et al., 1997). The results of two experiments are reported. As expected, antisaccade error rates were increased in accordance with the fronto-executive demands of the n-back task (Experiment 1). In addition, the debilitating effects of working-memory load were restricted to the inhibitory component of the antisaccade task (Experiment 2). These findings corroborate the view that working memory operations play a critical role in the suppression of prepotent behavioral responses.

scholarly journals Learning to suppress a distractor is not affected by working memory load

2019 ◽  
Vol 27 (1) ◽  
pp. 96-104 ◽  
Author(s):  
Ya Gao ◽  
Jan Theeuwes
Keyword(s):  
Working Memory ◽  
High Probability ◽  
Spatial Working Memory ◽  
Memory Load ◽  
Working Memory Capacity ◽  
Memory Task ◽  
Automatic Process ◽  
Long Term Memory ◽  
Working Memory Load ◽  
Statistical Regularities

AbstractWhere and what we attend to is not only determined by our current goals but also by what we have encountered in the past. Recent studies have shown that people learn to extract statistical regularities in the environment resulting in attentional suppression of high-probability distractor locations, effectively reducing capture by a distractor. Here, we asked whether this statistical learning is dependent on working memory resources. The additional singleton task in which one location was more likely to contain a distractor was combined with a concurrent visual working memory task (Experiment 1) and a spatial working memory task (Experiment 2). The result showed that learning to suppress this high-probability location was not at all affected by working memory load. We conclude that learning to suppress a location is an implicit and automatic process that does not rely on visual or spatial working memory capacity, nor on executive control resources. We speculate that extracting regularities from the environment likely relies on long-term memory processes.

Performance Dissociation during Verbal and Spatial Working Memory Tasks

2007 ◽  
Vol 105 (1) ◽  
pp. 243-250 ◽  
Author(s):  
Bonnie J. Nagel ◽  
Arthur Ohannessian ◽  
Kevin Cummins
Keyword(s):  
Working Memory ◽  
Spatial Working Memory ◽  
Memory Load ◽  
Memory Task ◽  
Past Research ◽  
Neural Substrates ◽  
Individual Performance ◽  
Task Design ◽  
Brain Response ◽  
Working Memory Load

Past research has inconsistently distinguished the neural substrates of various types of working memory. Task design and individual performance differences are known to alter patterns of brain response during working-memory tasks. These task and individual differences may have produced discrepancies in imaging findings. This study of 50 healthy adults ( Mage = 19.6 yr., SD = .8) examined performance during various parametric manipulations of a verbal and spatial n-back working-memory task. Performance systematically dissociated on the basis of working-memory load, working memory type, and stimulus difficulty, with participants having greater accuracy but slower response time during conditions requiring verbal versus spatial working memory. These findings hold implications for cognitive and neuroimaging studies of verbal and spatial working memory and highlight the importance of considering both task design and individual behavior.

scholarly journals Three simple steps to improve the interpretability of EEG-SVM studies

2021 ◽  
Author(s):  
Coralie Joucla ◽  
Damien Gabriel ◽  
Emmanuel Haffen ◽  
Juan-Pablo Ortega
Keyword(s):  
Machine Learning ◽  
Model Development ◽  
Research Literature ◽  
Machine Learning Algorithms ◽  
Support Vector ◽  
Machine Learning Classification ◽  
Diagnosis And Prognosis ◽  
Eeg Data ◽  
Clinical Adoption

Research in machine-learning classification of electroencephalography (EEG) data offers important perspectives for the diagnosis and prognosis of a wide variety of neurological and psychiatric conditions, but the clinical adoption of such systems remains low. We propose here that much of the difficulties translating EEG-machine learning research to the clinic result from consistent inaccuracies in their technical reporting, which severely impair the interpretability of their often-high claims of performance. Taking example from a major class of machine-learning algorithms used in EEG research, the support-vector machine (SVM), we highlight three important aspects of model development (normalization, hyperparameter optimization and cross-validation) and show that, while these 3 aspects can make or break the performance of the system, they are left entirely undocumented in a shockingly vast majority of the research literature. Providing a more systematic description of these aspects of model development constitute three simple steps to improve the interpretability of EEG-SVM research and, in fine, its clinical adoption.

scholarly journals The Role of Cognitive Load in Intentional Forgetting Using the Think/No-Think Task

2017 ◽  
Vol 64 (1) ◽  
pp. 14-26 ◽  
Author(s):  
Saima Noreen ◽  
Jan W. de Fockert
Keyword(s):  
Working Memory ◽  
Memory Load ◽  
Memory Task ◽  
Load Condition ◽  
Cognitive Resources ◽  
Working Memory Load ◽  
Intentional Forgetting ◽  
Probe Test ◽  
Target Words

Abstract. We investigated the role of cognitive control in intentional forgetting by manipulating working memory load during the think/no-think task. In two experiments, participants learned a series of cue-target word pairs and were asked to recall the target words associated with some cues or to avoid thinking about the target associated with other cues. In addition to this, participants also performed a modified version of the n-back task which required them to respond to the identity of a single target letter present in the currently presented cue word (n = 0 condition, low working memory load), and in either the previous cue word (n = 1 condition, high working memory load, Experiment 1) or the cue word presented two trials previously (n = 2 condition, high working memory load, Experiment 2). Participants’ memory for the target words was subsequently tested using same and novel independent probes. In both experiments it was found that although participants were successful at forgetting on both the same and independent-probe tests in the low working memory load condition, they were only successful at forgetting on the same-probe test in the high working memory load condition. We argue that our findings suggest that the high load working memory task diverted attention from direct suppression and acted as an interference-based strategy. Thus, when cognitive resources are limited participants can switch between the strategies they use to prevent unwanted memories from coming to mind.

Sign in / Sign up

Export Citation Format

Share Document