Pjax

New frameworks offer gradient-free and hierarchical learning for stable deep network training

Two new research papers propose alternative methods for training deep neural networks. One paper introduces a projection-based framework called PJAX, which treats training as a feasibility problem solvable through itera…