Electroencephalograph data training system, brainwaves predicting system, and electroencephalograph data training method

Abstract

A brainwave data training system, a brainwave data training method, and a brainwave prediction system are disclosed. The brainwave data training method is executed by a processing module and includes: receiving multiple training data sets, each training data set containing a training tensor and label data; multiple elements of the training tensor on a first axis corresponding to multiple detection units of the brainwave instrument; and multiple elements of the training tensor on a second axis corresponding to multiple time points; and training a neural network module based on the training data to obtain post-training parameter values ​​for each parameter of the neural network model.

Claims

1. A brainwave data training system, wherein, The system includes: a processing module configured to receive multiple training data sets, each training data set comprising a training tensor and labeled data, multiple elements on a first axis of the training tensor corresponding to multiple detection units of an EEG device, and multiple elements on a second axis of the training tensor corresponding to multiple time points; and a neural network module comprising an initial neural network model and multiple parameters; the neural network module is configured to generate a neural network module output based on the neural network module input, the initial neural network model, and the parameters in response to receiving a neural network module input; and the processing module is configured to train the neural network module based on the training data to obtain a trained parameter value for each parameter.

2. The EEG data training system as described in claim 1, wherein, The neural network module includes at least one convolutional module and an output module, wherein the at least one convolutional module is configured to receive the neural network module input and generate a feature tensor, and the output module is configured to receive the feature tensor and generate the neural network module output corresponding to the feature tensor.

3. The brainwave data training system as described in claim 2, wherein, Each of the at least one convolutional module comprises a one-dimensional convolutional layer, a rectified linear unit layer, and a one-dimensional pooling layer.

4. The brainwave data training system as described in claim 2, wherein, The output module includes a flattening module, a fully connected layer, and a normalized exponent layer.

5. The brainwave data training system as described in claim 1, wherein, The neural network module includes a recursive neural network module and an output module, wherein the recursive neural network module is configured to receive inputs to the neural network module and generate a recursive neural network module output; the output module is configured to receive the recursive neural network module output and generate a neural network module output corresponding to the recursive neural network module output.

6. The brainwave data training system as described in claim 5, wherein, The recursive neural network module includes at least one long short-term memory layer; the output module includes a fully connected layer and a normalized exponential layer.

7. The brainwave data training system as described in claim 1, wherein, The neural network module includes a convolutional module, a recursive neural network module, and an output module. The convolutional module is configured to receive the input of the neural network module and generate a feature tensor. The recursive neural network module is configured to generate a recursive neural network module output based on the feature tensor. The output module is configured to receive the recursive neural network module output and generate a neural network module output corresponding to the recursive neural network module output.

8. The brainwave data training system as described in claim 7, wherein, The convolutional module includes a one-dimensional convolutional layer, a rectified linear unit layer, and a one-dimensional pooling layer; the recursive neural network module includes at least one long short-term memory layer; and the output module includes a fully connected layer and a normalized exponential layer.

9. A brainwave prediction system, wherein the parameters are trained using a brainwave data training system as described in any one of claims 1 to 8, wherein, The brainwave prediction system includes: a prediction module comprising a prediction neural network model configured to have the same architecture as the initial neural network model, the prediction module being configured to store the trained parameter values ​​for each parameter and configured to generate a prediction output for the prediction input based on the prediction input, the prediction neural network model, and the trained parameter values ​​for each parameter in response to receiving a prediction input; and a prediction processing module configured to receive multiple time series in a time interval from multiple detection units of a prediction brainwave instrument, and generate a prediction tensor as the prediction input based on the time series to obtain the prediction output, wherein multiple elements of the prediction tensor on a first axis of the prediction tensor are the time series, and multiple elements of the prediction tensor on a second axis of the prediction tensor correspond to multiple time points within the time interval.

10. A method for training EEG data, used to train a neural network module, executed by a processing module; the neural network module includes an initial neural network model and multiple parameters; the neural network module is configured to generate a neural network module output based on the neural network module input, the initial neural network model, and the parameters in response to receiving a neural network module input; wherein, The training method includes: (a) receiving multiple training data, each of which includes a training tensor and label data, multiple elements on a first axis of the training tensor corresponding to multiple detection units of an EEG, and multiple elements on a second axis of the training tensor corresponding to multiple time points; and (b) training the neural network module based on the training data to obtain a post-training parameter value for each parameter.

Images