Convolutions vs Transformers

The cluster centers on debates about whether transformers or new ML architectures are essentially rediscovering or equivalent to convolutional neural networks (CNNs) and convolutions.

📉 Falling 0.4x AI & Machine Learning
2,370
Comments
18
Years Active
5
Top Authors
#3554
Topic ID

Activity Over Time

2009
1
2010
2
2011
3
2012
24
2013
31
2014
71
2015
151
2016
264
2017
362
2018
236
2019
224
2020
222
2021
164
2022
132
2023
206
2024
132
2025
138
2026
9

Keywords

C99 VGG16 youtube.com CNN distill.pub mlinpl.org ImageNet ConvNet ConvNets AMA neural cnn image networks neural network neural networks layers network nets images

Sample Comments

skdotdan Apr 24, 2019 View on HN

Are Transformers based on convolutions?

CMCDragonkai May 23, 2019 View on HN

Could this be used for convolutional neural networks?

wardedVibe May 27, 2022 View on HN

I see we are re-deriving conv nets

steinvakt2 Apr 17, 2025 View on HN

But isn't this basically what the conv layer does...?

smus Apr 27, 2025 View on HN

Convolutional neural networks are pretty big

davesque Dec 16, 2018 View on HN

Pretty broad but, from one section in the article, sounds like they're using convolutional layers in their networks?

CyberDildonics Dec 3, 2023 View on HN

It's just as bad a "convolutional neural networks" instead of "images being scaled down"

uoaei Nov 24, 2020 View on HN

Convolution is fancy linear regression

natch Oct 6, 2022 View on HN

They work at the level of convolutions, not images.

sorenjan Aug 6, 2023 View on HN

What's wrong about using the term convolution in ML?