“> Now we can learn even the image priors (spatial locality and translation invariance) from data. Right. The attention layers even learn attention patterns which look like convolution layer kernels! But better, presumably: https://arxiv.org/abs/1911.03584”
On the Relationship between Self-Attention and Convolutional Layers
Recent trends of incorporating attention mechanisms in vision have led researchers to reconsider the supremacy of convolutional layers as a primary building block. Beyond helping CNNs to handle long-range dependencies, Ramachandran et al. (2019) showed that attention can completely replace convolution and achieve state-of-the-art performance on vision tasks. This raises the question: do learned attention layers operate similarly to convolutional layers? This work provides evidence that attention layers can perform convolution and, indeed, they often learn to do so in practice. Specifically, we prove that a multi-head self-attention layer with sufficient number of heads is at least as expressive as any convolutional layer. Our numerical experiments then show that self-attention layers attend to pixel-grid patterns similarly to CNN layers, corroborating our analysis. Our code is publicly available.
候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“You could even argue that ML does classical vision in addition to other stuff. CNNs learn gabor filters. The AlexNet paper even shows this [0] Or if you look at the work ViT built itself on, they show attention heads will also learn these fillers. [1] That's actually a big part of how ViTs work, the heads integrate this type of information [0] https://…”
“It's worth noting that many of those projects were the famous versions of papers, ones that were successful through scale rather than through innovation. I'll give a good example, here's essentially the same architecture as the 16x16 words ViT paper but a year earlier[0]. It's not even the first, they even mention two other works that used transformers on…”