On the Relationship between Self-Attention and Convolutional Layers · HackerPapers discussions

On the Relationship between Self-Attention and Convolutional Layers — mentioned by godelski

Tue, 02 Sep 2025 02:47:34 GMT

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 images. I'm all for scaling and it is important, but there's tons of papers like this that are greatly overshadowed because someone just scaled up and got a better benchmark. It's been getting out of hand... Fuck... I'm starting to sound…

Paper and discussion history. A mention is not an endorsement.

On the Relationship between Self-Attention and Convolutional Layers — mentioned by godelski

Sat, 07 Jun 2025 03:41:37 GMT

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://papers.nips.cc/paper_files/paper/2012/hash/c399862d3... [1] https://arxiv.org/abs/1911.03584

Paper and discussion history. A mention is not an endorsement.

On the Relationship between Self-Attention and Convolutional Layers — mentioned by gwern

Wed, 17 Jun 2020 19:26:14 GMT

> 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

Paper and discussion history. A mention is not an endorsement.