For astronomical images, proceed to hyper-parameters tunning of ICA.
- On the model of performance tests executed with Keras, generate a set of transformations (with ground truth) for each class of transformations.
- Errors can be calculated between estimates and Ground truth.
- Hyper-parameter tunning can be done with cross-validation for example.
- Assess which robust error function to use in presence of noise and obstruction.
For astronomical images, proceed to hyper-parameters tunning of ICA.