Skip to content
ML Atlas

Deep Learning Architectures

How neural networks are wired for different data: convolutions for images, recurrence for sequences, attention and Transformers for nearly everything.

Role · kind
Domain
Level
Intermediate

All houses are made of bricks, but a lighthouse, a bridge and a barn are built very differently because they do different jobs. Architectures are different ways to arrange the same neural bricks.

Where it fits

Deep Learning ArchitecturesAlphaFoldAlphaFoldAttentionAttentionConvolutional Neural NetworkConvolutional Neural NetworkGraph Neural NetworkGraph Neural NetworkMixture of ExpertsMixture of Experts

Not shown: 4 more kinds and parts. See the full list below.

+All relationships as a list

See it on the full map

Overview

The fundamentals tell you how a single neural network learns. Architectures decide how the neurons are connected, and that choice matters enormously.

  • CNNs slide small filters over an image, so a pattern learned in one corner is recognized everywhere.
  • RNNs and LSTMs read a sequence one step at a time, carrying a memory forward.
  • Attention lets every element look directly at every other element and decide what is relevant.
  • Transformers are built almost entirely from attention. Since 2017 they have taken over language, and increasingly vision, audio and more.

A rough history: CNNs won image recognition in 2012 (AlexNet), RNNs and LSTMs dominated text until about 2017, and Transformers have dominated since.

Everyday use cases

  • Case 01

    Picking the right network

    Images, audio, text and graphs each have architectures that fit their structure.