New training method speeds up neural networks on hard problems

Original: SoftServe: A Scalable Quasi-Newton Method for Deep Learning

Source: arxiv.org ↗