Fetching the paper…

SGD learning on neural networks: leap complexity and saddle-to-saddle dynamics · Around