When the [[domain]] is "continuous" (ie [[subset]] of some $\mathbb{R}^{D}$).
Typically [[discrete time|iterative]] [[algorithm]]s.
A continuous optimization algorithm used in [[machine learning]] is usually called a **fitting method**
[[zeroth order optimization]]
- [[natural selection]]
[[gradient based optimization]] (first order)
- [[backpropagation]] algorithm for computing gradients of [[artificial neural network]]s
[[Hessian-based optimization]] (second order)
# implementation
- [[autodiff]]
- [[efficient training]]
- [[ResNet#^resnet-training]]
- [[efficient deep learning]]
- [[efficient finetuning]]
- [[artificial neural network]]
[`torch.optim` documentation](https://pytorch.org/docs/stable/optim.html#module-torch.optim)
- can vary per-layer learning rates by passing in list of `{ "params": [], ...keywords }` as the learning rate
- can adjust [[learning rate]] over time
https://github.com/ClashLuke/TrueGrad
[[Optax]] has good list
# sources
[[2018HeathScientificComputing]]