ERRORS · LANDSCAPES · LEARNING Loss, What does a model get wrong? Turn mistakes into a number, then see the landscape that number creates.
Start reading → Mean squared error over weight and bias Invented linear example. Selected weight 0.5, bias 2.5, mean loss 3.750. The highlighted surface curve holds bias fixed. Diamond marks weight2,bias1,loss0. View angle 45 degrees affects projection only. Mean squared error over weight and bias Invented linear example. Selected weight 0.5, bias 2.5, mean loss 3.750. The highlighted surface curve holds bias fixed. Diamond marks weight2,bias1,loss0. View angle 45 degrees affects projection only. Two model parameters. One height: mean squared error. An exact landscape for three invented points.
Five short chapters, from a single error to a loss surface. Start with basic ML and arithmetic; we introduce derivatives and log loss along the way. No model download required.
01 Errors into a number A prediction can miss in many ways. A loss function decides how those misses count.
↗ 02 Different penalties The same residual can receive very different penalties. That choice changes what improvement means.
↗ 03 A landscape of choices Two parameters choose the predictions. Their mean loss becomes the height of a surface.
↗ 04 Following the slope A gradient describes local change. A learning rate decides how far to trust that direction.
↗ 05 Probabilities and purpose A loss can reward better probability assignments without matching a thresholded score—or proving generalization.
↗