Deep learning
딥러닝
A branch of machine learning that stacks many layers of artificial neurons.
With enough layers, early ones pick up edges and simple shapes while later ones capture faces or objects. Unlike earlier methods, the features are not hand designed.
The shift came in 2012, when a deep model won an image recognition contest by a wide margin (Krizhevsky et al., 2012). Speech, translation and generation followed.
The cost is data and compute, which is why most teams adapt an existing trained model rather than build one from scratch.
- What depth meansFeatures are abstracted stage by stage.
- Practical choiceLook at fine tuning or zero shot before training from scratch.