Neural Networks¶
Neural network-based methods use deep learning to learn the inverse mapping from sensor data to source activity. They require training data but can capture complex nonlinear relationships. Note: These solvers require TensorFlow.
This category contains 5 solvers.
Solvers¶
| Full Name | Solver ID | Description |
|---|---|---|
| Convolutional Neural Network | CNN |
Supervised CNN that maps sensor time series to source activity using simulated training data. |
| Covariance-based Convolutional Neural Network | CovCNN |
Supervised CNN that operates on sensor covariance features (optionally with shrinkage) to predict source activity on ... |
| CovCNN (KL divergence) | CovCNN-KL |
Supervised ANN on sensor covariance trained with KL divergence between predicted and true L1-normalized source distri... |
| Fully-Connected Neural Network | FC |
Supervised fully-connected network trained on simulated data to map sensor time series to source activity. |
| Long Short-Term Memory Network | LSTM |
Supervised recurrent (LSTM) network trained on simulated data to map sensor time series to source activity. |