Skip to content
The methodOpen data
Back to the collection

MLP-Mixer: An all-MLP Architecture for Vision

Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov

Show all 12 authors

Lucas Beyer, Xiaohua Zhai, Thomas Unterthiner, Jessica Yung, Andreas Steiner, Daniel Keysers, Jakob Uszkoreit, Mario Lucic, Alexey Dosovitskiy

2021 · arXiv

Computer visionLanguage modelsMachine learning
Read the paper
HN score
1
Mentions
3
Readers
3
Months
3
What’s in a score?

From the abstract

Convolutional Neural Networks (CNNs) are the go-to model for computer vision. Recently, attention-based networks, such as the Vision Transformer, have also become popular. In this paper we show that while convolutions and attention are both sufficient for good performance, neither of them are necessary. We present MLP-Mixer, an architecture based exclusively on multi-layer perceptrons (MLPs). MLP-Mixer contains two types of layers: one with MLPs applied independently to image patches (i.e. "mixing" the per-location features), and one with MLPs applied across patches (i.e. "mixing" spatial information). When trained on large datasets, or with modern regularization schemes, MLP-Mixer attains competitive scores on image classification benchmarks, with pre-training and inference cost comparable to state-of-the-art models. We hope that these results spark further research beyond the realms of well established CNNs and Transformers.

Metadata: arXiv. Abstracts retain their original authors’ rights.

In the conversation

Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels

3 selected excerpts · 3 total mentions

godelski
Recommendation · rule
> Also, we know that transformers can scale Do we have strong evidence that other models don't scale or have we just put more time into transformers? Convolutional resnets look to scale on vision and language: (cv) https://arxiv.org/abs/2301.00808, (cv) https://arxiv.org/abs/2110.00476, (nlp) https://github.com/HazyResearch/safari MLPs also seem to scale: (cv) https://arxiv.org/abs/2105.01601, (cv) https://arxiv.org/abs/2105.03404 I mean I don't see a strong reason to turn away from attention as well but I also don't think anyone's thrown a billion parameter MLP or Conv model at a problem. We've put a lot of work into attention, transformers, and scaling these. Thousands…
Read the full comment
of papers each year! Definitely don't see that for other architectures. The ResNet Strikes back paper is a great paper for one reason being that it should remind us all to not get lost in the hype and that our advancements are coupled. We learned a lot of training techniques since the original ResNet days and pushing those to ResNets also makes them a lot better and really closes the gaps. At least in vision (where I research). It is easy to railroad in research where we have publish or perish and hype driven reviewing.
Original comment MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers
mattalex
Neutral mention · rule
Because the size of the attention matrix depends on the number of tokens (this is what makes attention N^2). If you don't care about having a flexible number of input tokens (e.g. in image processing) you can learn a fixed routing matrix. This is known as an MLP mixer https://arxiv.org/pdf/2105.01601 : you have one layer that processes each token in isolation ("vertical MLP") but ignores the inter-token connections, followed by a layer that combines between tokens ("horizontal MLP") that treats the internals of every token identically.
Original comment Do transformers need three projections? Systematic study of QKV variants
gwern
Neutral mention · rule
Actually... we're not sure! While it's true that Transformers work amazingly on more domains than they have any right to, it's unclear if they truly obsolete multi-layer perceptrons/fully-connected networks. You may have seen the recent splash of MLP-Mixer for image classification using perceptrons (https://arxiv.org/abs/2105.01601#google) but FC-heavy or FC-only nets have popped up here and there, never quite going away, and slowly getting better initializations & residual layers making them ever more trainable. I have a bibliography of some links I've noted over the years about them: https://www.gwern.net/notes/FC I've wondered if MLPs will be another example of the Bitter Lesson - the additional flexibility and power was their undoing until data/compute caught up and 'grad student descent' figured out how to use them right... If in the next 5 years everyone gets on the "MLP Is All You Need" train, I will be only mildly surprised.
Original comment X-Transformers: A fully-featured transformer with experimental features

Keep following the thread.

More in Computer vision

More from the same topics.