LabRuns entirely in your browser. Nothing is sent to any server.
The ball that hunts the minimum
Three balls with different learning rates race to the valley of a loss surface. Too large a step and they overshoot; too small and they never arrive. All of neural network training is this.
More experiments: