# Dynamic [Sparsity](Sparsity.md) - train intrinsically sparse neural networks from scratch using only a small proportion of parameters and FLOPs - Dynamic [Sparsity](Sparsity.md) enables training sparse models from scratch, hence the training and inference FLOPs and memory requirements are only a small fraction of the dense models. - models built with dynamic [Sparsity](Sparsity.md) can be trained from scratch to match their dense counterparts without involving any pre-training or dense training