ICA
Extract statistically independent source signals from mixed observations
Independent Component Analysis (ICA) finds a linear transformation that makes the output components as statistically independent as possible. It is commonly used for blind source separation, such as separating individual audio sources from microphone recordings.
When to use:
- Signal separation (EEG, audio, sensor data with mixed sources)
- When statistical independence between components is more useful than variance maximization
- Feature extraction that captures distinct underlying generative processes
Input: Tabular data with the feature columns defined during training Output: Independent component activations for each row
Model Settings (set during training, used at inference)
N Components (default: 2) Number of independent components to extract.
Algorithm (default: parallel)
ICA algorithm. parallel updates all components simultaneously; deflation extracts them one by one.
Max Iterations (default: 200) Maximum iterations for convergence.
Tolerance (default: 0.0001) Convergence threshold.
Inference Settings
No dedicated inference-time settings. The trained unmixing matrix transforms new data into independent components.