| Feedforward (MLP) |
One way flow input to output |
Dense layers, activations |
Tabular, regression |
Baseline model |
| CNN |
Spatial feature extraction |
Convolution, pooling |
Images, video |
Translation invariant |
| RNN |
Sequence modeling with state |
Hidden state |
Time series, text |
Vanishing gradients |
| LSTM |
Gated long memory |
Input forget output gates |
Long sequences |
Stable training |
| Transformer |
Attention based modeling |
Self attention, multi head |
NLP, vision |
Highly parallel |
| Autoencoder |
Compression via reconstruction |
Encoder, decoder |
Anomaly detection |
Unsupervised |
| GAN |
Adversarial generation |
Generator, discriminator |
Image synthesis |
Training instability |
| GNN |
Graph message passing |
Node edge embeddings |
Networks, molecules |
Relational reasoning |
| PINN |
Physics constrained learning |
PDE residuals |
Scientific computing |
Data efficient |