David Kappel

State-space models can learn in-context by gradient descent

Asynchronous stochastic gradient descent with decoupled backpropagation and layer-wise updates

Block-local learning with probabilistic latent representations