A comparison of algorithms for fitting the PARAFAC model

摘要: A multitude of algorithms have been developed to fit a trilinear PARAFAC model three-way array. Limits and advantages some the available methods (i.e. GRAM-DTLD, PARAFAC-ALS, ASD, SWATLD, PMF3 dGN) are compared. The explained in general terms together with two approaches accelerate them: line search compression. In order compare different methods, 720 sets artificial data were generated varying level type noise, collinearity factors rank. Two models fitted on each set: first having correct number F second F+1 components (the objective being assess sensitivity over-factoring problem, i.e. when extracted exceeds rank array). also tested real fluorescence measurements, again by extracting both right an exceeding factors. evaluations based on: iterations necessary reach convergence, time consumption, quality solution amount resources required for calculations (primarily memory).