Computer Vision News - June 2016

14 Computer Vision News Tool Let’s explain those: (2) eigenvalue decomposition will be used for initialization; (4) Alternating Least Squares (ALS) will be used for the main algorithm; (6-7) setting the stopping criteria; and finally in (9) we decompose the tensor into 3 low rank matrix. This decomposition is illustrated in the following figure: In this data exploration step, let’s inspect vector A13 of Factor1 which can also be named the ‘subject score’. Note that in CPD (similarly to PCA) the first factor explains the largest portion of the variance of the data. The A13 reveals the following picture: With the provided slider, one can scroll through the images and have a first taste of it. Next, we will decompose the above tensor T. We will start by setting the various parameters for the decomposition process.