convergence acceleration
Convergence acceleration refers to techniques employed to speed up the convergence of optimization algorithms in training neural networks, allowing the model to reach an optimal solution faster than standard approaches.
- A Bayesian Fast-Slow Framework to Mitigate Interference in Non-Stationary Reinforcement Learning
- Decreasing Entropic Regularization Averaged Gradient for Semi-Discrete Optimal Transport
- FastDINOv2: Frequency Based Curriculum Learning Improves Robustness and Training Speed
- Flick: Empowering Federated Learning with Commonsense Knowledge
- Improving Energy Natural Gradient Descent through Woodbury, Momentum, and Randomization
- Moment- and Power-Spectrum-Based Gaussianity Regularization for Text-to-Image Models
- ORIGEN: Zero-Shot 3D Orientation Grounding in Text-to-Image Generation
- Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think
- Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think
- Second-order Optimization under Heavy-Tailed Noise: Hessian Clipping and Sample Complexity Limits