product network: the binding units would have r subscripts: ti1i2...ir ; there would be r sides; there would be r input/output vectors, say u1 ,u2 ,. . . ,ur ; to teach the tensor the fact u1 (u2 , . . . , ur ), add u1i1 × u2i2 × . . . × urir to each binding unit ti1i2...ir ; to retrieve, say, the r-th component given the ﬁrst r − 1, you would compute urir = i1,i2,...,ir−1 ti1i2...ir u1i1 u2i2 . . . urir−1 Network Topology Tensor Product Nets Relational Memories High Ranks TPN Applications 28/40