I want to beat backpropagation. How it's going: memory almost on par, wall-clock time 7x while converging to same result. I'm optimizing a narrow task first (MLP on MNIST data) and I'm using the Evolution Strategies algorithm EGGROLL with some adaptations. In the video I show the progress I'm making.
johann_zahlmann•1h ago