… This paper proposes a new quasi-attention that is compositional in nature, ie, learning whether
… Our proposed\textit {Compositional De-Attention}(CoDA) is fundamentally built upon the …

… Next, we propose an improved Quasi-Attention CG-BERT model that learns a compositional
attention that supports subtractive attention. We train both models with pretrained BERT on …

… We employ a compositional attention [22] mechanism for this purpose. Self-attention [26,35]
is initially designed for capturing long-range dependencies. However, it learns to allocate …

Attention Model has now become an important concept in neural networks that has been … We
also describe how attention has been used to improve the interpretability of neural networks…

… -networks, we employ compositional de-attention mechanism [… networks via a mutual
information maximization strategy. We also propose to leverage compositional deattention network …

… We also transfer compositional de-attention (CoDA) to solve the unique keypoint occlusion
problem in this task and propose a novel keypoint feature replacement to recover the extreme …

… 2019) as our backbone network as it is the best-performing baseline (see Table 4). … All such
compositionally manipulated vectors ZC ij , after interacting with each jth description vectors …

… Compositional De-Attention is generally better for refining prototypes than scaled cosine
attention, which supports our hypothesis that CoDA has more representation learning capacity …

… Inspired by the format of the Compositional De-Attention [14], which imports the dissimilarity
representation as to the extra information, we adopt a similar expression to apply the …

… [32], we use a Compositional De-Attention to fuse the information from the two sets. Given
the sets of disentangled representations, the Compositional DeAttention computes an attentive …