MVICAD2: Multi-View Independent Component Analysis with Delays and Dilations
| Source: Apple ML Research
Tags: Apple ML Research, ICA, MEG, neuroscience, multi-view learning, biomedical AI, signal processing
MVICAD2 extends multi-view ICA for MEG brain data by adding dilation handling alongside delays — critical for neuroscience studies where brain processing speed varies by age and stimulus type, validated on the Cam-CAN aging dataset.
Details
Magnetoencephalography (MEG) records magnetic fields from neural activity at millisecond resolution, making it valuable for studying brain dynamics. When analyzing responses to auditory stimuli across participants of different ages, two problems arise: different brain regions respond at different latencies (delays), and older brains may process the same stimulus at a slower timescale overall (dilations). Standard ICA ignores both. MVICAD (the predecessor from Apple ML Research) handled delays between views but not dilations. MVICAD2 adds dilation parameters to the model, allowing each view to independently stretch or compress its temporal representation before fitting the shared independent components. This is particularly important for auditory paradigms where processing speed is known to decrease with age. The method is validated on the Cam-CAN dataset, a large-scale multi-site MEG/MRI study of cognitive aging across the adult lifespan. The paper is published in IEEE Transactions on Biomedical Engineering. The practical impact is most direct for researchers running multi-participant MEG studies where age is a variable of interest, but the dilation extension generalizes to any multi-view ICA scenario with known temporal scaling variation.