Non-negative factorization revealed common factors of representation dynamics among different reach direction. a. Tensor arrangement before decomposition. Three dimensions were reach direction, a merged time component, and a merged muscle & electrode component. b. Weights of the reach direction dimension, respectively for each of the four factors. c. Weights of the time dimension, respectively for each of the four factors.

Non-negative factorization revealed common factors of representation dynamics among different reach direction. a. Tensor arrangement before decomposition. Three dimensions were reach direction, a merged time component, and a merged muscle & electrode component. b. Weights of the reach direction dimension, respectively for each of the four factors. c. Weights of the time dimension, respectively for each of the four factors.

Source publication
Preprint
Full-text available
The primary motor cortex (M1) is known to be spatially organized in terms of muscles and body parts, notwithstanding some overlap. However, the classical view is largely static, neglecting potential representational changes on a fast time scale. To address potential representational dynamics, we probed the mutual information between M1 signals reco...

Contexts in source publication

Context 1
... we aimed to simplify the understanding of dynamic representation for different reach directions, muscles and sites. For this purpose, we performed non-negative matrix decomposition on the tensor of mutual information (Fig. 6 a, see also methods for details), and showed that the representation for different movements could be understood using a unified set of bases. We performed the decomposition using 2~5 factors, which produced relative errors of 0.6261, 0.6169, 0.6164, 0.6120 respectively. We chose 4 factors, considering both interpretability and ...
Context 2
... using a unified set of bases. We performed the decomposition using 2~5 factors, which produced relative errors of 0.6261, 0.6169, 0.6164, 0.6120 respectively. We chose 4 factors, considering both interpretability and faithfulness of reconstruction. Weights of the dimensions of reach direction were quite continuous for adjacent directions (Fig. 6 b). Representation for rightward and inward movements had later peaks and was slightly noisier in general (Fig. 6 c factor 2 and ...
Context 3
... of 0.6261, 0.6169, 0.6164, 0.6120 respectively. We chose 4 factors, considering both interpretability and faithfulness of reconstruction. Weights of the dimensions of reach direction were quite continuous for adjacent directions (Fig. 6 b). Representation for rightward and inward movements had later peaks and was slightly noisier in general (Fig. 6 c factor 2 and ...