Figure 6. Experimental results for decentralized joint independent component analysis (djICA) ( Baker et al., 2015).

The experiment is based on synthetic functional MRI data using a generalized autoregressive conditional heteroscedastic model ( Engle, 1982; Bollerslev, 1986). The top figure shows that as the global number of subjects increases, the Moreau-Amari index (MAI) decreases for both pooled-data ICA and djICA with different principal component analysis (PCA) operations. Additionally, MAI converges for pooled-data ICA and djICA when the number of subjects increases. The bottom figure shows that number of splits in the data have no effect on MAI.