Apple researchers advance brain signal analysis with MVICAD2 technique

Multi-view machine learning—analyzing data from multiple sources simultaneously—has long struggled with integration of heterogeneous data and alignment of feature spaces across subjects. In neuroscience, this challenge is particularly acute: group-level studies of brain activity typically assume that neural sources are identical across all participants, an assumption the body identifies as often too restrictive given individual variability and age-related changes. Earlier techniques like Multi-View Independent Component Analysis and Multi-View Independent Component Analysis with Delays made progress by relaxing some constraints, but temporal dilation effects—variations in the speed at which neural responses unfold—remained unaddressed. MVICAD2 fills this gap by jointly modeling both delays and dilations, supported by a mathematically identifiable model with a closed-form likelihood approximation. The validation on the Cam-CAN dataset, showing correlation between delays/dilations and aging, suggests that accounting for these variations is not merely a technical improvement but clinically and scientifically meaningful.