Search papers, labs, and topics across Lattice.
3
0
4
Training dynamics of Transformers can be reduced to a low-dimensional manifold, revealing how inductive reasoning emerges from data statistics and model initialization.
The learning rate is redefined as a structural parameter of training dynamics, fundamentally shaping the representations selected by gradient descent.
The Oracle Distance theorem reveals that all noising processes achieve the same optimal negative ELBO, linking diverse loss functions in diffusion models.
