Code for this paper: http://bbs.pinggu.org/forum.php?mod=viewthread&tid=6214243&pid=49150963&page=1should be coupled to capsule j(b)The log priors can be learned discriminatively at the same time as all the other weights. They dependon the location and type of the two capsules but not on the current input image. The initial couplingcoefficients are then iteratively refined by measuring the agreement between the current output v, ofeach capsule, j, in the layer above and the prediction u,i made by capsule iThe agreement is simply the scalar product aij=V;uili. This agreement is treated as if it was a lolikelihood and is added to the initial logit, bij before computing the new values for all the couplingcoefficients linking capsule i to higher level capsulesIn convolutional capsule layers, each capsule outputs a local grid of vectors to each type of capsule inthe layer above using different transformation matrices for each member of the grid as well as foreach type of capsuleProcedure 1 Routing algorithm1: procedure ROUTING(uji,T,2: for all capsule i in layer I and capsule j in layer(L+1): bij<03: for r iterations dofor all capsule i in layer 1:Ci+softmax(bi)D sof tmax computes Eqfor all capsule j in layer(I+1: Si+for all capsule j in layer(+1: v;