exponential moving average
A method for smoothing data over time to provide a more stable estimate of model parameters such as weights during training and inference.
- Dense Backpropagation Improves Training for Sparse Mixture-of-Experts
- E-BATS: Efficient Backpropagation-Free Test-Time Adaptation for Speech Foundation Models
- Orthogonal Contrastive Learning for Multi-Representation fMRI Analysis
- PseuZO: Pseudo-Zeroth-Order Algorithm for Training Deep Neural Networks
- Robust Neural Rendering in the Wild with Asymmetric Dual 3D Gaussian Splatting