Independent generation of sequence elements by motor cortex; rnn-sequence-model

Overview

This project contains the trained RNN model that was used to generate Fig. 7-8 of Zimnik and Churchland (2020). With the attached code, one is able to load the trained model, generate the inputs that the model was trained on (as well as the target outputs), feed the inputs to the model, and observe the resulting output (i.e., muscle activity).

Code

This project contains a single code file, stored in the 'Code' directory: a Jupyter Notebook titled 'Sequence_RNN_Model.ipynb'.

Data

This project contains two data files, stored in the 'Input Data' directory: 'SequenceModel', which contains the trained RNN and 'SequenceModel_data' which contains empirical preparatory activity and empirical EMG, which are used to produce the model input and target output, respectively.

Use

Users are able to change the 'instructed pause' (i.e., duration of time between the required bouts of muscle activity) by changing the value of the variable 'IP' in the appropriate cell (the appropriate cell is marked in the Notebook). Doing so demonstrates that the model is able to produce both delayed double-reaches and compound reaches. Note: users should not change any parameters other than 'IP'.

Please contact the authors to get access to the files.

Citation

Andrew J. Zimnik, Mark M. Churchland (2021): Independent generation of sequence elements by motor cortex; rnn-sequence-model, Gigantum, Inc https://doi.org/10.34747/n3rs-9w57

Associated Publication

Andrew J. Zimnik, Mark M. Churchland (2021): Independent generation of sequence elements by motor cortex, Springer Science and Business Media LLC https://doi.org/10.1038/s41593-021-00798-5