Independent component analysis using an extended infomax algorithm for mixed subgaussian and supergaussian sources

摘要: An extension of the infomax algorithm Bell and Sejnowski (1995) is presented that able blindly to separate mixed signals with sub- supergaussian source distributions. This was achieved by using a simple type learning rule first derived Girolami (1997) choosing negentropy as projection pursuit index. Parameterized probability distributions have regimes were used derive general preserves architecture proposed (1995), optimized natural gradient Amari (1998), uses stability analysis Cardoso Laheld (1996) switch between regimes. We demonstrate extended 20 sources variety easily. Applied high-dimensional data from electroencephalographic recordings, it effective at separating artifacts such eye blinks line noise weaker electrical arise sou...