A data-driven approach to decompose motion data into task-relevant and task-irrelevant components in categorical outcome

Abstract

In this paper, we propose a data-driven technique to detect task-relevant and task-irrelevant motion components with categorical task outcomes. Our method relies on a linear regression technique for solving classification problems, such as logistic regression 7. For example, logistic regression enables classification of whether the current motion data are associated with throwing a fastball or breaking ball. Our data-driven method can be applied even when the relation between motion data and outcome is unknown: the relation can still be estimated in a data-driven manner. Along with recent data-driven approaches in biomechanics that focused on unsupervised methods 8,9 , we rely on supervised methods to address the task-relevant and task-irrelevant components with categorical outcomes in a data-driven manner. Notably, our current mathematical framework is the same as that used in our previous methods 5,6 , which focused on continuous task outcomes by applying a linear regression technique. Along with our previous methods, we propose a unified data-driven approach to detect task-relevant and task-irrelevant motion components for multiple kinds of task outcomes. Methods participants. Eight healthy volunteers (aged 18-21 years, four females) participated in our experiment for two days (not consecutively). All the participants were informed of the experimental procedures and their conformance with the Declaration of Helsinki, and all participants provided written informed consent before the start of the experiments. All the procedures were approved by the Ethics Committee of the Tokyo University of Agriculture and Technology.

Effort-dependent effects on uniform and diverse muscle activity features in skilled pitching

“…Another perspective on the redundancy problem is the decomposition of motion data into task-relevant and task-irrelevant components 28 , 29 and the suppression of motor variability, especially in task-relevant motion components 29 . To discuss task-relevant and task-irrelevant features, different types of data-driven methods 30 – 32 are preferable rather than tensor decomposition. Because the current study focused on the effort-dependent effects on multiple and time-varying muscle activities, we relied on tensor decomposition.…”

Detecting task-relevant spatiotemporal modules and their relation to motor adaptation

“….,K). To estimate the relevance, our earlier studies demonstrated the effectiveness of a ridge regression rather than some nonlinear regression techniques [10,23,24]. A ridge regression allows us to estimate the relevance of motion to performance data W2R [I, J] while minimizing prediction error in the presence of observation noise (see Methods for details).…”

“…A comparison of motion-relevant and task-relevant modules provides insight into the relationship between motion-relevant modules and task-relevant components. First, our framework is based on ridge regression [22] to predict task performance based on time-varying multidimensional motion data [10,23,24]. A ridge regression allows us to quantify how motion data in each time frame in each body part are relevant to predicting task performance.…”

“…In our study, we calculated relevant modules to predict the task outcome (the lower part). We assume that a linear relationship exists between motion and performance (the lower-left part), which has been validated in previous studies [10,23,24]. The bracket in the lower-left part indicates the inner product of the matrices W and X :,:,k .…”

Impaired feedforward control of movements in pianists with focal dystonia

“…We thus evaluated the motion components relevant to the classification of tempo into its fastest and natural components while expecting to quantify the effects of skill learning on feedforward control. To extract the nontrivial classification-relevant motion components, we utilized a data-driven technique (21)(22)(23).…”