Reviews: Deep Multimodal Multilinear Fusion with High-order Polynomial Pooling

The preliminaries section lays out the required mathematical formulations of tensor networks. While the section presented serves its function, it could potentially be more clear if the authors spaced out and typeset the maths (similar to how it is done in Section 3.1). The visualizations in Figure 2/3 illustrates how a fusion network (and hierarchical network) can be constructed with PTP units. These visualizations clearly communicate how features are pooled across modality and time step. That said, perhaps the descriptions about the HPFN (Section 3.2) are overly verbose.