← PreviousYou are hereReadPartial derivatives and the gradientOn a surface over two parameters, take the slope along each axis with the other held still; the two slopes form the gradient, which points straight uphill.
Next · Linear regression and gradient descentYou are hereReadGradient descentMake a loss small by stepping against its gradient, and see the learning rate decide whether the steps crawl, glide, zig-zag or fly off.