A multilinear subspace regression method: HOPLS

 

Matlab code of HOPLS 



Reference: Q. Zhao, C. F. Caiafa, D. P. Mandic, Z. C. Chao, Y. Nagasaka, N. Fujii, L. Zhang and A. Cichocki. Higher-Order Partial Least Squares (HOPLS): A Generalized Multi-linear Regression Method. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(7):1660-1673, 2013

DownloadHOPLS_files/2012HOPLS_Demo.rar

Decoding of behavior data from brain data

The video below illustrates the excellent prediction performance of HOPLS, by comparing the actual and predicted 3D movement trajectories of multiple markers based on ECoG signals recorded from the monkey brain.  The actual hand movement trajectory (green line) and the trajectory predicted by HOPLS (red line) are very close, despite the large number of degrees of freedom in the 3D space of multiple markers. The predicted points refer to the most recent half-second window and  the corresponding variations of beta- and gamma-band ECoG power spectrum are  shown on the right, according to the 32 channel convention and the usual intensity coding (red - high magnitude, blue - low magnitude).